col-xs-12
col-sm-12
col-md-12
col-lg-12
col-xs-12
col-sm-12
col-md-12
col-lg-12
videoType
dynamic-media
videoDmUrl
https://delivery-p136806-e1377785.adobeaemcloud.com/adobe/assets/urn:aaid:aem:6f7beddd-4d82-4ab5-b6b4-5a23b63e2f05/play?assetname=AllanHsiaoPalmOil.mp4
keyFrameImage
timestamp

00:01 for palm oil and in substitutes.

00:04 Now that's that first issue.

00:06 Second,

00:06 there's going to be this commitment problem,

00:08 and here's where the dynamics kick in.

00:10 OK,

00:10 so

00:11 once those forests are cut down,

00:13 ex post,

00:13 there will now be this temptation to reduce tariffs.

00:16 Why?

00:16 Because after the forest is gone,

00:18 those emissions are sunk.

00:20 But on the supply side,

00:22 farmers are thinking about their future palm oil revenues.

00:25 Again,

00:25 so then the problem is this.

00:26 Without commitment,

00:28 future tariffs are low.

00:29 And so if I'm a farmer,

00:30 I'm just going to keep on deforesting today,

00:32 even if

00:33 today's tariff happens to be high.

00:35 Why?

00:35 Because I'm thinking about the future mostly.

00:38 So,

00:39 the dynamics are really going to matter

00:41 economically here.

00:42 Uh,

00:42 they're also going to turn out to matter quantitatively.

00:45 But dynamics,

00:45 of course,

00:46 are going to make estimation much harder computationally.

00:49 So I'm also going to make a small methodological contribution here,

00:52 building on existing oiler techniques to

00:55 overcome

00:56 this challenge.

00:57 And then in terms of data,

00:58 satellite data are going to show me these

01:00 palm oil plantations and mills both over space and

01:04 where they're being developed over time.

01:07 So to preview results a little bit,

01:09 I'm gonna find that import tariffs are effective

01:11 relative to the domestic palm oil tax.

01:14 If it's the case that importers are coordinating

01:17 and they can commit to long-term policy.

01:20 Ah,

01:20 but here,

01:21 coordination and commitment are compliments,

01:23 OK,

01:23 and so we really need both of them.

01:25 If we don't have coordination,

01:27 we're going to have low coverage,

01:29 which leads to low,

01:30 uh,

01:31 which leads to leakage and therefore to low tariffs.

01:34 And then without commitment we can't uphold tariffs and

01:37 so we're going to have low tariffs tomorrow.

01:40 We therefore want to compensate with high tariffs today,

01:43 but we can't under leakage because high tariffs are

01:45 just pushing palm oil over to unregulated markets.

01:49 So then what do we end up with,

01:50 we end up with low tariffs tomorrow in addition to low tariffs today.

01:55 So then to touch briefly on contributions,

01:57 I have this new dynamic empirical framework for evaluating

02:01 emission-based trade policy,

02:03 and what I want to highlight here is

02:05 that by focusing on one particular important industry,

02:09 I'm going to have a pretty rich model of supply

02:12 that's going to capture very important features like the dynamics

02:15 and like the spatial heterogene.

02:17 Now,

02:18 second,

02:19 leakage and commitment are independently well

02:21 recognized and decently well studied,

02:23 of course,

02:24 in the very well studied,

02:25 of course,

02:26 in the environmental literature.

02:28 Uh,

02:28 I'm gonna combine them within a single framework and in doing so,

02:31 be able to show how they interact.

02:33 And that interaction is new,

02:35 especially in an empirical setting.

02:38 And of course speaking of empirics,

02:39 I'm going to have these empirical estimates for palm oil,

02:42 which is understudied,

02:44 but and I don't want to understate this,

02:46 a massive huge source of emissions

02:48 that is important in and of itself

02:50 if we want to hit our climate targets.

02:54 So

02:54 just to outline,

02:55 I'll start with the setting.

02:56 Then we'll go to the demand model which speaks to leakage,

02:59 then the supply model,

03:01 which is dynamic and speaks to commitment.

03:04 And then for supply,

03:05 let me highlight that the structural model captures

03:08 that these investment decisions are forward looking.

03:11 OK,

03:11 and so future prices matter here.

03:13 It's not going to be enough

03:14 for us to just regress palm oil supply today on the price today.

03:19 That's not all that farmers are thinking about.

03:21 And then last,

03:22 I'll run counterfactuals to quantify leakage

03:24 and commitment.

03:26 So maybe let me just spend 2 minutes going,

03:28 uh,

03:28 giving you highlights on the setting,

03:30 and then I'll pause for questions.

03:32 So,

03:32 palm oil is a perennial crop.

03:34 Once you grow the tree,

03:35 it's going to keep giving you this fruit,

03:37 OK?

03:37 And this is what the fruit looks like.

03:39 Uh,

03:39 that fruit is going to be taken to a nearby mill,

03:42 and it's going to become palm oil.

03:43 It's going to get milled into palm oil.

03:45 Of course,

03:46 when we set up plantations to grow that palm oil,

03:48 we're deforesting and destroying forest,

03:51 and in this region,

03:52 a lot of that forest is actually a special kind of forest called peatland

03:57 forest.

03:57 So here's a before and after,

03:58 left

03:59 and right.

04:00 Peatland forests are forests that sit on swamps that contain

04:04 deep layers of decomposing organic material

04:07 called peat,

04:08 and the trees would recover in about 100 years,

04:10 but that peat would take 10,000 years

04:12 to recover,

04:13 very much non-renewable.

04:15 And furthermore,

04:16 that peat has 10 times more carbon than the actual trees themselves.

04:21 Slash and burn is releasing all of that carbon into the air,

04:24 and that's why palm oil emissions are so,

04:27 so big

04:27 and so significant.

04:29 From there,

04:30 palm oil is going to be used in many familiar products,

04:32 both food like

04:34 cookies and margarine,

04:35 as well as non-food,

04:36 detergents,

04:37 soaps,

04:38 and even cosmetics like lipstick.

04:40 Now who's producing palm oil?

04:41 It's Indonesia and Malaysia.

04:43 They're producing a lot of palm oil

04:45 and they're doing it for export.

04:47 So they're 84% of world production

04:49 and 90% of exports.

04:52 For world consumption,

04:53 Indonesia and Malaysia consume 20% domestically.

04:56 And then

04:57 the top three importers,

04:58 the EU,

04:59 China,

04:59 and India,

05:00 consume another 35%.

05:02 So

05:03 notice here that China and India

05:06 don't really produce any,

05:07 don't produce any palm oil themselves.

05:10 So import tariffs here are really just a consumption tax with

05:14 no discrimination needed based on where

05:17 this palm oil is coming from.

05:19 And just to wrap up this kind of settings section

05:23 here I'm showing you how this industry is expanding over 3 decades over time.

05:28 So the dark blue here of course is palm oil plantations,

05:32 and what we see is as time unfolds,

05:36 you know,

05:36 as things unfold,

05:37 we're seeing this massive expansion

05:39 over time and over space,

05:42 and all of that land sort of

05:44 devastatingly is getting deforested.

05:46 All of that peatland is going up into the air.

05:48 In the form of emissions.

05:50 So what am I doing with uh counterfactuals?

05:52 I'm gonna roll back to 1988,

05:54 put in regulation in the form of import tariffs,

05:57 and then

05:57 I'm going to roll forward

05:59 and see

06:00 how things unfold

06:02 and change in response.

06:04 So let me pause briefly for,

06:06 for questions before we talk about the demand estimation.

06:15 Great,

06:16 if,

06:16 if there are no questions,

06:17 I'll just jump into the,

06:19 this is,

06:19 uh,

06:19 this is Tristan,

06:20 sorry,

06:20 I was waiting for the moderator,

06:21 but I'll go.

06:22 Uh,

06:23 so,

06:23 uh,

06:24 Article 2 of the Paris Agreement

06:26 recognizes that countries have differentiated responsibilities

06:30 and capabilities,

06:32 uh,

06:32 in addressing

06:34 climate change,

06:35 um,

06:36 and you know this case of Indonesia is one where clearly,

06:39 you know,

06:39 you have a,

06:40 a lower income country producing something that's carbon intensive.

06:44 And this is a policy of

06:46 rich countries,

06:48 uh,

06:48 to stop that.

06:49 Um,

06:50 I,

06:50 I guess how,

06:51 how would you account

06:52 for

06:53 an idea of differentiated responsibility,

06:56 ah,

06:57 in this model,

06:58 uh,

06:58 and you know,

06:59 is there something you can say about that?

07:01 Yes,

07:02 thank you for the question.

07:03 So

07:03 differentiated

07:04 ability is certainly

07:07 Part and parcel of the approach here.

07:08 This is kind of the motivation for thinking about these external external

07:12 policies like importers as opposed to

07:14 relying on the Indonesian government itself,

07:16 which may have enforcement issues,

07:18 uh,

07:19 you know,

07:19 at play.

07:20 However,

07:20 the differentiated responsibility,

07:22 ultimately I'm,

07:23 I will be able to speak to a bit.

07:25 In particular,

07:26 you're getting at this equity concern,

07:27 right,

07:28 where Indonesia and Malaysia happen to be

07:30 endowed with certain natural resources that are potentially theirs to exploit,

07:35 uh,

07:35 particularly Indonesia as a lower income country.

07:37 How can we be sensitive to that?

07:39 At the very end,

07:40 I'm going to show you

07:42 what those equity concerns.

07:43 Look like specifically because I have the full supply side specified,

07:47 so I can quantify that producer surplus loss for Indonesia,

07:51 for Malaysia

07:52 in response to any kind of tariff that we put in place in counterfactuals.

07:56 That kind of producer surplus loss is the kind of

07:59 loss,

08:00 the kind of damages that one would want to address

08:03 with

08:03 transfers in order to address this equity.

08:06 Um,

08:07 but you know,

08:08 and one can also think about waiting,

08:10 right?

08:10 So when tariffs are set,

08:12 they're set

08:12 in my baseline analysis,

08:14 waiting,

08:15 you know,

08:15 a dollar of producer surplus in Indonesia,

08:17 1 to 1

08:19 compared to consumer surplus in Europe,

08:21 which is maybe.

08:22 Unilevers $1 of consumer surplus,

08:24 OK,

08:25 uh,

08:25 but if you want to upweight Indonesian surplus

08:27 because maybe that's lifting people out of poverty,

08:30 then in that case,

08:31 maybe you want to upweight

08:33 producer surplus,

08:34 and,

08:34 and that would change the kind of tariff that is,

08:36 that is optimal.

08:39 Hi,

08:40 uh,

08:40 this is Govindra Timilna.

08:42 So it looks like you have a single commodity model,

08:44 right?

08:44 It's only for

08:45 palm oil,

08:47 but palm oil is used basically,

08:48 you know,

08:49 to produce the biodiesel.

08:51 For example,

08:51 even if there is an import tariff for palm oil,

08:55 Indonesia,

08:55 Malaysian farm,

08:56 they can produce biodiesel,

08:58 and they can export biodiesel,

09:00 right?

09:01 So in this case,

09:01 the leakage still would be there.

09:04 Yes,

09:04 so,

09:04 so the simple policy response is

09:07 to also put

09:08 the carbon tax on biodiesel,

09:10 right?

09:10 So not just palm oil,

09:12 raw palm oil,

09:12 but rather

09:13 palm oil content.

09:15 And,

09:16 and so that's going to prevent that margin of leakage.

09:19 Um,

09:19 but the,

09:19 the other thing I'll point out,

09:20 and we'll see this in demand is,

09:22 yes,

09:22 this is one commodity,

09:23 of course,

09:24 but I'm not thinking about it purely in isolation.

09:26 Right,

09:27 because on the European side,

09:28 for example,

09:29 uh,

09:29 palm oil is going to be a substitute with soybean oil,

09:32 rapeseed oil,

09:33 both of which the European Union produces a lot of,

09:36 particularly rapeseed oil.

09:38 And so demand,

09:39 of course,

09:40 is going to need to account for that,

09:41 right?

09:41 It's,

09:41 it's not as if palm oil is the one and only good

09:44 that people

09:45 use and,

09:46 and can't substitute to anything else.

09:50 Go ahead.

09:53 Please,

09:55 please.

09:56 Yes,

09:56 so,

09:57 uh,

09:58 regarding the single commodity model,

10:00 Indonesia clear field not just for palm oil.

10:04 They clear field for all other crops as well.

10:07 And how do you account for that?

10:08 Are you going to say all the deforestation is all due to palm oil plantation,

10:14 or there's some other way

10:16 that you're going to measure

10:17 specific for palm oil plantation?

10:20 Thank you.

10:21 So I,

10:22 to be very upfront,

10:23 I am not accounting for that in the

10:27 kind of estimates that I'll show you today.

10:28 And indeed this is going to be a source of carbon bias because,

10:31 you know,

10:32 you put in

10:32 palm oil tariffs,

10:33 palm oil production goes down,

10:35 but if that land is still being slashed and burned for acacia,

10:39 that's the number 2 crop in Indonesia,

10:41 then there's going to be huge leakage on that margin,

10:43 and then my

10:44 carbon estimates will be,

10:45 will be wrong.

10:46 They'll be biased.

10:47 Now the The reason I didn't put that into the model

10:50 explicitly is because I can actually check that in the data.

10:52 So I have data on acacia plantations,

10:55 and I can look at substitution from palm oil to acacia,

10:58 and that substitution turned out to be much smaller

11:00 than I expected.

11:01 I thought I was going to have a nested model where first

11:04 we choose as a farmer between all of these different crops.

11:07 Then if I choose palm,

11:08 it's this exact model.

11:09 If it's Acacia,

11:10 then within that nest,

11:11 it's again my exact model,

11:13 and I thought it was going to have this cross nest substitution,

11:15 but that substitution ended up being small,

11:17 and the reason I think is that

11:19 those other crops are all much less profitable

11:21 than palm oil.

11:22 The reason why basically all of the

11:24 deforestation in Indonesia and Malaysia is happening

11:26 for palm oil is because everyone is growing palm

11:29 oil because it's the most profitable crop by far.

11:31 And so part of this is just that,

11:33 you know,

11:34 Acacia is 7 times less profitable than palm oil.

11:37 That means that not everyone who's no longer

11:39 producing palm oil would move to Acacia,

11:41 and so that's why that substitution,

11:43 I measured to be small.

11:46 Otherwise I could have had this more complicated model,

11:48 but the substitution was small in the data,

11:49 so I just didn't,

11:50 didn't have that margin.

11:52 Uh,

11:53 I,

11:53 I,

11:54 I'm Malaysian.

11:55 I must say that I'm very surprised that you

11:58 associate all those,

12:00 uh,

12:00 development in Malaysia to

12:02 palm oil plantation,

12:04 which I know is going down,

12:05 not going up in Malaysia.

12:07 But this is what you,

12:08 you show on the map,

12:09 which is what

12:10 much of my surprise.

12:13 Yes,

12:14 so in,

12:14 in Malaysia,

12:14 well,

12:15 what's happened is it's been capped,

12:17 um,

12:17 and so the,

12:17 the Malaysian government says you can't,

12:19 you can't do more.

12:20 Of course,

12:20 Malaysia has,

12:21 you know,

12:22 it has more ability to enforce than than the Indonesian government.

12:26 Um,

12:27 but then also,

12:27 you know,

12:28 you can,

12:28 you can put a cap,

12:29 but then once you cut down the forest,

12:30 it's all gone already anyway,

12:31 right?

12:31 So it's,

12:32 it's,

12:32 uh,

12:32 there's no kind of

12:34 going back on the peatland that,

12:35 that has been destroyed.

12:37 Um,

12:37 but yes,

12:37 I,

12:38 I thought that there was going to be much more substitution.

12:40 It,

12:41 it turned out not to be,

12:42 um,

12:43 but that's,

12:43 you know,

12:44 That's in the data that that I observed.

12:46 In some counterfactual worlds,

12:48 maybe,

12:48 you know,

12:49 uh,

12:50 you don't have the economies of scale in palm oil,

12:51 and now Acacia takes off,

12:53 uh,

12:53 in that world,

12:54 then I'm not going to quite be capturing all of that.

12:56 Um,

12:57 for that though,

12:58 I would need a model

12:59 of

12:59 the Acacia industry,

13:00 which isn't quite what I have here.

13:03 But,

13:03 but that's how we would,

13:04 that's what you would need is this multi-industry.

13:07 And again,

13:08 I measured the substitution to be small.

13:11 OK,

13:11 so I'll just jump into demand.

13:13 So

13:14 for demand,

13:15 I have an almost ideal demand system.

13:17 I'm going to move through relatively quickly here just because I want to

13:20 flag that the richness really is going to be on the supply side.

13:23 But nonetheless,

13:24 let's jump in.

13:25 This is typical two-stage budgeting.

13:27 First,

13:27 consumers are choosing how much to spend on the vegetable oil category.

13:31 That's that top line

13:32 with the log log specification.

13:34 And then second,

13:35 consumers are allocating that spending between palm oil and other vegetable oils.

13:40 I want to flag here that these substitutes,

13:42 and this point came up a little bit before,

13:44 these substitutes are not associated with big carbon emissions.

13:48 So I'm not that worried about carbon bias,

13:50 bias in my carbon estimates from this margin of leakage.

13:54 Now the asterisk here is that South American soybean oil.

13:57 does have some deforestation associated with it,

13:59 with it,

14:00 but I think that that bias is going to be small even there because one,

14:03 it's only South American soybean oil.

14:05 Second,

14:06 Amazonian deforestation is really driven more directly by cattle than soybeans.

14:10 And third,

14:11 it's not peatland destruction.

14:13 So we're also talking about an order of magnitude,

14:15 order of magnitude less

14:17 emissions than what we're talking about with palm oil.

14:20 So then on the second line,

14:21 omega is the palm oil expenditure share,

14:24 Omega IT.

14:25 It's a function of 1,

14:27 a sort of secular oil specific time trend,

14:29 2,

14:30 the prices of each oil product.

14:32 OK,

14:32 so I'm going to get on price elasticities as well as

14:35 cross price elasticities,

14:37 and then 3,

14:38 the category budget.

14:40 So what I want to flag here is that this is a product

14:42 space demand system that's going to allow me to be very flexible on

14:46 the cross-product substitution patterns.

14:48 Now,

14:49 capturing that substitution is precisely the advantage

14:51 here relative to a reduced form approach,

14:54 where,

14:54 for example,

14:54 if I just regressed palm oil demand

14:57 on prices,

14:57 even of course if I use instruments,

14:59 then I wouldn't be fully accounting

15:01 for that switching to these alternative products

15:03 and therefore the demand elasticities that I would get would be,

15:06 would be biased.

15:07 So again,

15:08 to think about leakage,

15:09 I want those market specific demand elasticities.

15:12 I'm going to get them by estimating

15:15 demand by market.

15:15 And then in terms of data,

15:16 I have annual consumption by oil product and country.

15:20 And then annual prices by oil.

15:23 So of course you know prices are endogenous,

15:25 price endogenity on the demand side.

15:26 What do I want?

15:27 I want a supply shifter.

15:29 For that supply shifter,

15:30 I'm going to use growing season rainfall shocks for foreign oil producers.

15:33 For example,

15:34 I'm going to instrument for US palm oil prices with rainfall shocks in Indonesia,

15:39 which produces palm oil.

15:41 So the bottom line here

15:42 in terms of results is going to be that the

15:45 importers,

15:46 that is the dark blue lines,

15:47 have relatively elastic demand.

15:50 OK,

15:50 and so what this means is that leakage is a problem.

15:53 If China is unregulated,

15:54 then because Chinese demand

15:56 is relatively elastic,

15:57 it means that it is going to expand and therefore it will offset

16:01 EU tariffs.

16:01 OK,

16:02 leakage is an issue.

16:03 On the other hand,

16:04 actually,

16:05 if importers coordinate,

16:06 then leakage

16:07 is not so bad.

16:09 Leakage is not so bad because

16:12 what do import tariffs kind of always miss,

16:14 even if importers all coordinate?

16:16 They're going to be missing domestic consumption in Indonesia and Malaysia,

16:19 which

16:20 isn't exported.

16:21 However,

16:22 this domestic demand turns out.

16:23 To be

16:24 pretty inelastic.

16:25 That's the light blue line.

16:26 And,

16:27 and so leakage on that margin isn't that bad.

16:29 And here I think it's,

16:30 it's inelastic

16:31 just because

16:32 Indonesia and Malaysia produce so much palm oil that really palm oil is the main

16:36 thing that is used.

16:37 They're not really using too much olive oil,

16:39 and so there's just less substitution available

16:42 and therefore a lower

16:43 demand elasticity.

16:45 And then I can also aggregate up to give you the,

16:49 I can also aggregate up to give you the world demand curve.

16:51 So the 2015 curve is going to be farther to the right than the 1990 curve,

16:56 that is to say,

16:57 more demand

16:57 at all prices,

16:59 and I can trace out that intercept to give

17:01 you the path of the rightward shift over time.

17:04 And that's the graph on the right.

17:05 So demand is rising as palm oil gets widely adopted,

17:09 and note that this is in logs.

17:11 So demand is rising

17:12 fast

17:13 over this period of time.

17:15 This is why plantations have expanded so much and so quickly,

17:19 as we saw on the map before.

17:21 This is also why without regulation,

17:23 because demand is going up and that's putting upward pressure on prices,

17:26 this is why we would continue to keep expanding in the absence of regulation.

17:32 Uh,

17:32 so that's an answer.

17:34 I,

17:34 I take a quick pause for questions.

17:35 Yes,

17:36 yes,

17:37 uh,

17:37 David McKenzie,

17:38 please go ahead and then Tom,

17:40 uh,

17:40 yes,

17:40 so,

17:40 so this was,

17:41 uh,

17:41 uh,

17:41 just a quick clarification.

17:42 When it's,

17:43 when you're looking at the domestic consumption,

17:45 is that picking up then,

17:47 um,

17:47 the use of palm oil as an intermediate input

17:50 into these other products?

17:51 And so the reason that it's inelastic may be because of the

17:55 sort of export of

17:56 some of these,

17:57 these other products that are using palm oil.

18:00 Yes,

18:00 in this,

18:01 in,

18:01 in terms of the data,

18:02 this USDA data,

18:03 it's,

18:03 it's measuring all of the consumption of the raw good,

18:06 whether it be used ah directly ah for a final consumer product,

18:10 as a final consumer product,

18:12 or whether it be used as an input to other,

18:14 other

18:15 industries.

18:16 Yes.

18:22 Thanks for that.

18:22 Please proceed.

18:24 Great.

18:25 Um,

18:25 so,

18:26 I will jump now into

18:27 supply.

18:29 So for supply,

18:30 I have this dynamic model

18:31 with sunk investment.

18:33 So I'm going to divide land into sites,

18:36 OK,

18:36 and a site is a block of land that can invest in palm oil.

18:40 And empty sites then are going to be potential entrants.

18:42 OK,

18:42 this is how I'm defining potential entrants in this market,

18:45 and note that,

18:45 you know,

18:46 you can only have so many potential entrants

18:48 because there's only so much land in Indonesia and Malaysia.

18:52 Now,

18:52 active sites are going to be operating,

18:54 they're gonna be producing palm oil,

18:56 they're gonna have one mill

18:57 and some set of plantations that are up,

19:00 up and running and online.

19:02 So this is going to be a Hoppenheim style

19:03 entry investment game with a dynamic competitive equilibrium.

19:07 That is to say that sites are going to invest in

19:09 deforest today in order to set up their palm oil trees,

19:12 and then those trees are going to produce revenues in every future period.

19:16 Now there's no exit here because,

19:18 you know,

19:18 once you've already paid the cost to set up the tree,

19:21 marginal costs at that point are pretty low,

19:23 so there's not really a big reason to exit anyway.

19:25 Also,

19:26 even if you exit,

19:26 deforestation has sunk at that point,

19:28 and that's really what I care about.

19:30 And then where do tariffs enter?

19:32 What's the point of tariffs?

19:33 Well,

19:33 tariffs reduce future revenues,

19:35 and that in turn is going to reduce the incentive

19:37 to invest and deforest in the first place today.

19:41 Now,

19:42 the structural model here,

19:43 what is it buying for us?

19:44 What is it capturing?

19:45 It's capturing that future prices matter because these are forward looking

19:49 investment decisions.

19:50 So again,

19:51 we don't just want to regress palm oil supply today on

19:53 prices today because prices today are not the one and only thing

19:56 that farmers are basing their investment decisions on.

20:00 And furthermore,

20:00 these are future prices,

20:01 so it's really expected future prices that matter,

20:05 not realized future prices,

20:07 and the structural model is going to help

20:08 me deal with those expectations in a flexible way

20:12 and kind of specify the assumptions that I need to make on them.

20:15 Now,

20:16 in terms of the timeline,

20:17 let me just show it to you graphically.

20:19 This timeline is gonna summarize the choices.

20:21 OK,

20:21 again,

20:21 we have the discrete and continuous elements here.

20:24 So,

20:25 so what are they?

20:26 Well,

20:26 we are an

20:27 empty site,

20:28 OK,

20:28 that's,

20:28 let's start on the left.

20:30 Uh,

20:30 we're gonna choose whether to build a mill or not.

20:33 That's a discrete yes or no binary choice.

20:36 If no,

20:37 then we're in the lower branch.

20:38 OK,

20:38 and we're gonna have the same choice next period.

20:41 If yes,

20:42 then we're in an upper branch,

20:43 in the upper branch,

20:44 and then what we're going to do is we're going to pay that

20:47 mill investment cost because we've chosen yes to build the mill.

20:51 Then once you are there and you've built that mill,

20:55 then you get to choose how much land to develop into plantations.

20:59 On the intensive margin,

21:00 and that's going to be a continuous choice.

21:02 Should we develop 456,

21:05 6.5,

21:06 you know,

21:06 6.25,

21:07 whatever

21:08 hectares of land into plantation.

21:10 And of course we're going to pay that plantation investment cost.

21:12 Big plantation,

21:14 lots of land,

21:14 bigger cost.

21:15 That's

21:17 now,

21:17 you know,

21:18 what's the point of having all of this investment costs that we're paying?

21:22 Well,

21:22 now starting in the next period,

21:24 we're going to be

21:24 producing palm oil and getting revenue.

21:27 And then at a later point,

21:28 we can also choose to expand our plantation.

21:31 We don't need to develop all of the land at once.

21:34 So a few comments here.

21:35 First,

21:36 these costs are going to be subject to logit cost shocks

21:40 that we're going to realize before we make the decision,

21:42 and we're going to base the decision on.

21:44 Uh,

21:45 Second,

21:45 again,

21:45 these are future revenues,

21:46 so we need to be dealing with expectations.

21:49 And then third,

21:51 these two margins of spatial dependence,

21:54 uh,

21:54 these two margins are going to capture that spatial dependence

21:57 between a mill and the land around it,

21:59 OK,

21:59 and,

22:00 and also that big fixed cost

22:01 for building the mill in the first place.

22:03 So it's not like every hectare of land has

22:05 its own cost function and its own cost shock.

22:08 No,

22:09 you know,

22:09 there are groups of land that are working together as one unit,

22:12 that's what these two margins are going to help capture.

22:16 Now for the state variables,

22:17 we have aggregate supply and aggregate demand which are observed

22:21 and of course they're going to affect world prices,

22:23 which in turn affect the revenues that firms are going

22:26 to get and therefore affect the incentive to invest.

22:29 So for supply,

22:30 very simply,

22:31 you know,

22:31 many plantations online,

22:33 high supply,

22:34 low prices.

22:35 And then for demand,

22:36 demand curve far out to the right,

22:38 high demand,

22:39 high prices.

22:40 So

22:41 in terms of spatial he

22:43 heterogeneity,

22:44 which is very important here,

22:45 sites are going to be spatially different in observed and unobserved ways.

22:51 So there's observed heterogeneity here in palm oil yields,

22:54 how much you can grow on any given plot of land,

22:56 and also cost factors like distance to a port,

22:58 which are important,

22:59 a major port,

23:00 which are important for

23:01 transport costs.

23:03 So that is to say that sites can be observably different.

23:06 And why is this important?

23:07 This is important because it's going to capture directly

23:10 that

23:11 this idea of the good land being developed first and therefore

23:15 the supply elasticity of the land that is left is going

23:17 to depend on what kind of land that is left,

23:20 and probably it's the worst land if it's,

23:22 it's

23:23 the kind of land that's been left

23:25 for a later period to develop.

23:27 But there's a restriction here,

23:28 OK,

23:28 the restriction is that there is no unobserved

23:32 site level heterogeneity

23:34 above and beyond the logic cost shocks,

23:36 which,

23:36 you know,

23:37 that's heterogeneity,

23:38 of course,

23:38 but it has a distributional assumption on it,

23:40 so it's restricted.

23:42 There's no site level unobserved heterogeneity,

23:44 like a,

23:45 a site fixed effect,

23:46 if you want to think of it that way,

23:48 because in this model,

23:49 once a mill is built,

23:50 it's built.

23:51 So I don't observe

23:52 this site level decision again and again over time.

23:56 And so I can't identify that site level fixed that.

23:59 Uh,

23:59 Instead,

24:00 what I need to do is I need to aggregate across space instead of over just over time.

24:04 And therefore I'm restricting unobserved heterogeneity here

24:07 to enter only at the regional level.

24:10 Uh,

24:10 so what that means is that I'm gonna

24:11 have regional fixed effects and regional time trends.

24:14 Uh,

24:14 but let me note that that's still

24:16 going to capture some very important unobservables,

24:19 like,

24:19 for example,

24:20 political economy,

24:22 where,

24:22 you know,

24:22 plantations in certain corrupt regions might be

24:25 cheaper because the bribes are cheaper there,

24:27 for example.

24:28 I don't see that,

24:29 but that's going to be absorbed

24:30 by this

24:31 regional unobserved heterogeny.

24:34 And then in terms of data,

24:35 I'm going to see these plantations and these mills,

24:39 these plantation and mill investments over time

24:41 and over space in the satellite data.

24:44 I'm also going to see revenues.

24:46 So I see world prices over time,

24:48 and I'm going to have quantities from the yields data.

24:50 So for yields,

24:51 I'm using an agronomy model developed specifically for palm oil.

24:55 This is a Hoffman et al.

24:56 2014,

24:58 and then I'm going to be able to feed

25:00 in climate data and get much more fine-grained than,

25:02 for example,

25:03 the FAO's agricultural suitability data,

25:06 which,

25:06 you know,

25:06 many people use,

25:07 but that's not quite what I'm using because it's not fine-grained enough

25:10 for,

25:10 for my purposes.

25:12 Now,

25:12 what I want to point out here in terms of the

25:14 quantity decision is that here active plantations are just producing their yield

25:18 in every period.

25:19 So on the,

25:20 the,

25:21 the quantity decision,

25:22 that is to say,

25:22 isn't

25:23 a matter of how much to produce on this given plot of land.

25:27 Instead,

25:28 that given plot of land is just producing its yield.

25:30 OK,

25:30 so the quantity decision instead is

25:32 over how much land to develop in the first place,

25:35 and again,

25:35 I see that by satellite.

25:38 And also

25:39 cost factors that vary over space.

25:41 So again,

25:41 you know,

25:42 the port distance thing,

25:43 but also being close to major roads

25:46 or major urban areas might be better because it lowers transport costs.

25:49 Uh,

25:50 and then I'm also gonna

25:51 estimate how carbon stocks enter firms' cost functions.

25:55 So what I'm gonna show you in a bit is that firms don't find

25:58 emissions privately costly.

26:00 That is to say that without regulation,

26:02 they're just gonna keep on emitting because they don't care

26:04 about those emissions.

26:05 It doesn't enter their cost function.

26:08 Uh it's a question by.

26:10 Please go ahead,

26:11 Mr.

26:11 Alan.

26:12 So the question is

26:14 why,

26:15 so I don't know if the government does it,

26:16 but wouldn't it be natural for the Indonesian government to levy an export tax

26:21 just to extract monopoly rents?

26:24 Yes,

26:24 so I,

26:24 I am going to show you that counterfactual at the very end actually of the export tax.

26:28 Ah,

26:28 maybe we think that it's administratively

26:31 easier to administer,

26:32 you know,

26:32 easier to administer because

26:34 now you can just impose it at the port

26:36 instead of,

26:37 you know,

26:37 1000 different auditors across all of the forests of Indonesia.

26:40 So I am going to show you that.

26:42 Yes,

26:43 Indonesia,

26:44 because it has market power,

26:45 will have this incentive to put in

26:47 some kind of tax to,

26:48 you know,

26:48 a monopolist wants to reduce supply in order to

26:50 drive up prices because they're a price maker.

26:53 It turns out that that motive will

26:55 impose some kind of tariff.

26:57 That tariff is not going to be nearly big enough

27:00 to match the size of the carbon externality that we're talking about today.

27:03 So yes,

27:04 that incentive is there,

27:05 but it's small relative to the carbon externality.

27:08 But then that tax is the entry point for the EU instead of trying to import.

27:13 Why don't you give an incentive for the government to increase the tax?

27:16 Yeah,

27:17 so the very last counterfactual,

27:18 I will show you this exact,

27:20 um,

27:21 this exact counterfactual,

27:22 and then I'll show you why

27:23 actually the,

27:24 the Indonesian government,

27:25 uh,

27:25 isn't happy with that,

27:27 that export tax as well.

27:28 Once I have the graph up,

27:29 it'll be a little easier to,

27:30 to,

27:31 to come to.

27:31 So,

27:32 so if I,

27:32 I,

27:32 I'll,

27:33 uh,

27:33 remember to bring this back up,

27:34 but if I don't,

27:35 please remind me.

27:35 It's on the last slide.

27:37 So,

27:37 uh,

27:38 Say also had a question.

27:40 Hi,

27:41 yes,

27:41 um,

27:41 so,

27:42 couple of things.

27:43 So,

27:43 uh,

27:44 first,

27:44 so should we,

27:45 you know,

27:46 most of the production then of palm oil in Indonesia is,

27:48 is,

27:49 uh,

27:49 you know,

27:49 individual meals,

27:51 or are there com you know,

27:52 multinational,

27:53 or are there companies basically that have multiple meals and so.

27:55 So,

27:56 so what,

27:56 what share of the production is,

27:58 is coming from,

27:59 you know,

28:00 individual,

28:00 individual mills.

28:01 And then,

28:02 so,

28:02 I mean,

28:02 I guess related to that,

28:03 I mean,

28:03 should we think about these producers as

28:06 price takers,

28:07 uh,

28:07 you know,

28:07 given that

28:08 most of the palm oil is produced in,

28:10 in Indonesia

28:11 and Malaysia,

28:12 again,

28:12 if,

28:13 if there's,

28:13 you know,

28:13 a large firm,

28:14 then,

28:14 you know,

28:15 the,

28:15 you know,

28:15 it's unclear that there would be.

28:17 Uh,

28:18 price takers,

28:19 and then,

28:20 you know,

28:20 sorry,

28:21 one last

28:21 point.

28:22 So,

28:23 so should we,

28:23 should we think about,

28:25 um,

28:25 so I guess,

28:26 so there's,

28:27 you know,

28:27 in terms of production,

28:28 there's no pests or other,

28:30 uh,

28:31 or,

28:31 or,

28:31 you know,

28:32 is that,

28:32 is that subsumed into,

28:34 into the,

28:34 into the climate shock,

28:35 I guess,

28:36 you know,

28:36 if there's,

28:36 you know,

28:37 if there's,

28:37 you know,

28:37 uh,

28:38 pests and stuff that might affect production,

28:40 um,

28:41 you know,

28:41 as well,

28:42 you know,

28:43 yeah.

28:44 Uh,

28:44 so is it the case that climate is going

28:46 to change production directly and are producers internalizing that?

28:50 Um,

28:51 that,

28:51 no,

28:52 because,

28:52 uh,

28:52 it's happening over such a long margin.

28:54 Also,

28:55 palm oil is pretty hardy a crop,

28:56 so temperatures can go up a bit and it's,

28:58 it,

28:58 it will actually be fine.

29:00 Um,

29:00 but then are there pests,

29:01 are there pests,

29:02 and,

29:03 you know,

29:03 uh,

29:04 that,

29:04 that,

29:04 that might,

29:05 might,

29:06 you know,

29:06 yeah,

29:07 might,

29:07 might.

29:09 You know,

29:09 that damage the production of palm oil.

29:12 There

29:13 probably there are some.

29:14 I'm sure I don't know the name of the specific pests that palm oil farmers deal with,

29:18 to be upfront.

29:20 So I don't,

29:20 I'm not controlling for any of that directly,

29:22 but that is going to be

29:24 picked up when I estimate the cost structure of the industry,

29:27 part of the costs of developing

29:29 a given plot of land into

29:31 plantation,

29:32 operational plantation is going to be inclusive of pesticide,

29:35 for example,

29:36 to deal with these pests.

29:37 Fair enough,

29:38 fair enough.

29:38 And then to get to your point on market power and market concentration.

29:43 There are

29:44 multinationals that have multiple mills.

29:46 There are about 1200 mills

29:48 in the data,

29:49 but you know,

29:50 um,

29:53 not Sinar Mas,

29:54 Wilmar

29:55 International is a huge company,

29:56 OK,

29:57 one of the biggest palm oil traders.

29:59 They actually only have 8 mills.

30:00 OK,

30:00 so they have 8,

30:01 it's more than 1,

30:02 but it's not

30:03 800.

30:04 And so I'm assuming everyone is a price taker,

30:06 and this is why

30:07 we see

30:08 that this market concentration here is pretty low.

30:11 So the,

30:12 the top.

30:12 One producer

30:14 in

30:15 the world is

30:16 FGV Burhad FGV Holdings Burhad.

30:20 This is a Malaysian company.

30:22 It's only 4% of world production.

30:24 The top

30:25 20 producers are 27%.

30:27 So it's a pretty unconcentrated market.

30:29 This is going to be important because I have a

30:32 substantive

30:33 assumption that I need to make to have estimation be tractable,

30:37 which rules out this,

30:38 this market power.

30:39 So that's,

30:39 that's also why I'm just assuming everyone is a price taker.

30:43 Um,

30:45 in terms of,

30:46 sorry,

30:46 sorry,

30:47 there are a couple more questions,

30:48 Alan,

30:48 uh,

30:48 Aaron,

30:49 then,

30:50 uh,

30:50 Ariana,

30:51 then Harris.

30:52 Try to both the questions and the answer is

30:55 a little brief in the interest of time.

30:58 Sure,

30:59 so I'm going to like ask very briefly.

31:00 So

31:01 I think in the upper branch of the decision tree,

31:03 you,

31:03 you observe like planting more or

31:06 like

31:06 not planting every year,

31:07 right?

31:08 If that's the case,

31:09 you can,

31:10 you can

31:11 estimate the unobserved heterogeneityity based on that like

31:13 the Harold Sucher's bus engine replacement problem,

31:16 right?

31:16 So is that correct?

31:18 That's correct.

31:18 But then the problem is going to be that I

31:20 don't see that until the mill is actually built.

31:23 So for many places,

31:24 I,

31:24 you know,

31:24 I have 30 years of data,

31:26 but if the mill didn't get built until 2005,

31:28 then actually I only have the tail end.

31:31 So I do see my decisions,

31:32 but it's not,

31:33 and it's not balanced across all of the,

31:35 all the mills.

31:36 And so that's,

31:37 that's why.

31:37 And then at the,

31:38 that's the intensive margin,

31:39 the upper branch,

31:40 but then for the mill itself,

31:42 it's you can't do it at all.

31:43 That's why I,

31:44 yeah,

31:44 that's why I just aggregated it to be regional for both.

31:48 Thanks.

31:50 I was wondering,

31:52 yeah,

31:52 I was wondering whether you can comment on on location specific production.

31:57 Uh,

31:57 so from what I understood,

31:58 from what I understand,

31:59 you said,

32:00 the issue is the specific,

32:02 uh,

32:03 environmental conditions in Malaysia and Indonesia.

32:05 And,

32:06 and so I'm,

32:06 I'm interested to understand whether the policy

32:08 should be targeting a reduction in those settings

32:10 and,

32:11 and potentially an increase in other settings such as

32:14 Africa,

32:15 Gabon,

32:15 and other places that are interested in,

32:17 uh,

32:17 in this area,

32:18 or should we view it simply as an

32:21 absolute reduction in production of.

32:25 Yes,

32:25 this is a very important point that you bring up.

32:27 I,

32:27 let me first say that I I am focusing on just

32:29 Indonesia and Malaysia because that's where I have very rich data,

32:32 and also they're 90% of production anyway,

32:35 so I just

32:36 call it a day there.

32:37 Uh,

32:37 but we can also think about kind of

32:39 supply leakage to West Africa,

32:41 South America.

32:42 There is some production there.

32:43 Nigeria,

32:43 for example,

32:44 is 3%

32:45 of world production.

32:46 But then

32:47 even within Indonesia,

32:49 there is this

32:50 question of whether all palm oil is equally bad,

32:53 and there the answer is no.

32:54 So palm oil produced on peatland forest is 10 times worse

32:57 than palm oil produced on non-peatland forest.

33:00 Not all forest in Indonesia

33:02 and Malaysia is peatland forest.

33:04 And so then what we realized then is that this import tariff that I'm talking about,

33:08 which treats

33:09 in my baseline,

33:10 I'm just going to treat all palm oil the same,

33:12 good and bad palm oil.

33:13 That's a blunt tool,

33:14 right?

33:14 Ideally,

33:15 and that's not a Peruvian thing.

33:16 Really,

33:17 the Peruvian thing would be to differentiate between the

33:19 good palm oil and the bad palm oil.

33:21 So I do run that counterfactual.

33:23 That is an important counterfactual.

33:24 I don't

33:25 set that as my baseline because when we're talking about import tariffs,

33:29 actually it's much easier to just treat all palm oil the same,

33:32 because otherwise,

33:32 You need to know for this specific unit of palm oil,

33:35 I need to trace back its production history over

33:38 time and figure out where it came from,

33:39 and I need to either do it myself or trust a certifier to do it correctly.

33:44 Also,

33:45 they're,

33:45 you know,

33:45 in the unregulated market,

33:47 good and bad palm oil is a perfect substitute.

33:49 So you have switching,

33:50 you know,

33:50 I'll take all of the good palm oil from China and send

33:52 it to Europe because Europe says they want good palm oil.

33:55 So it's not a kind of

33:56 fix everything,

33:57 but,

33:57 but of course it will be better because it's targeted,

33:59 and I do run that counterfactually.

34:03 OK,

34:03 final question for Harris,

34:04 and then I think we should

34:06 move on in the interest of time.

34:08 OK,

34:08 hi.

34:09 So I have a question regarding how firms access the land.

34:12 I mean,

34:12 how much do they pay in order to obtain a lease?

34:16 It's something that seems to be a little bit hidden in your model,

34:18 maybe through the production costs.

34:19 I mean there's no explicit

34:20 modeling of the price of the land.

34:22 I mean,

34:22 there's been a lot of debate.

34:24 About whether governments somehow squander the land by giving you very

34:28 favorable access to our producers.

34:31 So

34:31 isn't there a possibility of

34:33 having some policy through making land more expensive

34:37 for producers rather than tax imports or exports.

34:40 Yes,

34:41 absolutely.

34:41 So this would be a surprise.

34:42 Well,

34:42 I,

34:43 I hear actually two things that you're saying.

34:44 One is access to export markets.

34:47 There,

34:48 I'm just assuming that

34:49 there's direct access and the world price is the world price.

34:52 I'm abstracting away from that.

34:53 But then at the level of,

34:55 you know,

34:55 land cost and production costs,

34:58 that's all getting picked up.

35:00 when I estimate the cost structure of the industry,

35:03 one point that can be made is,

35:04 you know,

35:04 when I run my counterfactuals,

35:05 I'm holding that cost structure fixed.

35:08 OK,

35:08 so if in response to palm oil tariffs,

35:10 the Indonesian government is responding

35:12 by changing the cost structure,

35:13 I'm going to miss that.

35:14 Although I can't get that without a model

35:15 of how Indonesia kind of sets cost structure.

35:18 But then also sort of,

35:20 what about this policy of bidding up?

35:22 What about Indonesia just bidding up the cost of land,

35:25 for example,

35:26 and reducing

35:27 things in that way.

35:28 There again,

35:28 that's going to rely on an Indonesian government that is both willing and able

35:32 to implement

35:33 that policy.

35:34 So one,

35:35 does the Indonesian government actually want to do that

35:37 given that palm oil is their number one agricultural export

35:40 and would be their top export,

35:43 full stop,

35:43 if it weren't for the fact that Indonesia is also a big natural gas oil producer.

35:47 Um,

35:48 but 2,

35:49 could the Indonesian government actually do it?

35:51 So they could put in a fine

35:53 for building in forested areas.

35:54 The central government could do that,

35:56 but then really because it's a decentralized government,

35:59 actually it's going to come down to enforcement by the local mayor

36:02 who,

36:02 you know,

36:02 I can bribe.

36:03 Ah,

36:03 and so there's an enforcement issue.

36:05 This is why I'm focused on the demand side.

36:07 These supply side policies are traditionally what environmental regulation

36:11 has focused on,

36:12 and I'm just pointing out that there's

36:13 this one other qualitatively different kind of approach

36:16 that might help in cases where

36:18 all of that

36:19 domestic regulation is very difficult and messy.

36:23 Actually,

36:23 we have another policy lever that we can use.

36:26 OK,

36:27 thanks for the clarification.

36:28 Thank you.

36:29 Great,

36:30 so I'll just talk a bit now about estimation.

36:32 So,

36:32 for estimation,

36:33 what I'm doing here is I'm combining classic continuous oiler methods

36:37 with newer discrete oiler CCP methods,

36:40 and both are going to be using short-term perturbations.

36:43 So,

36:43 I'm comparing investing today,

36:45 which is what actually happens in the data,

36:47 to delaying investment by one period,

36:50 and by revealed preference,

36:51 uh,

36:52 that

36:52 delayed investment is suboptimal because I didn't actually do it,

36:56 and that delay is going to help me pin down the sub

36:59 sub-optimality of that delaying is going to help me pin down the payoff parameters.

37:03 So notice here in both cases,

37:05 the investment is up and running by period T + 2.

37:08 And so some under some assumptions,

37:10 the continuation values are going to align and difference out.

37:14 Ah,

37:14 so differencing out the continuation values helps us a lot

37:17 computationally because now I don't need to compute them.

37:20 And that's the usual challenge.

37:21 Also,

37:22 future expectations are getting differenced

37:24 out alongside the continuation values.

37:26 That is to say,

37:26 this is the sense in which I can accommodate those expectations

37:30 without needing to specify exactly what they are,

37:32 of course,

37:33 subject to an assumption that I'm going to

37:35 talk about in just

37:36 one second.

37:37 Um,

37:37 so,

37:37 what am I assuming?

37:38 I'm assuming first that site owners are actually choosing between today or delay.

37:44 That's gonna put some restrictions on,

37:46 for example,

37:46 property rights or credit constraints or uncertainty,

37:49 although I can speak to each of those a little bit.

37:52 Now

37:52 I'm also assuming that sites are atomistic,

37:54 and this gets at Xavier's comment from before,

37:57 where this is going to rule out market power,

37:59 OK,

38:00 but this is a place where this is a setting where market concentration is low,

38:03 maybe that's not that bad,

38:05 but it's also gonna rule out spatial interaction,

38:07 uh,

38:07 which,

38:08 you know,

38:08 if I am

38:10 a site that is neighboring you,

38:12 another site,

38:12 and now because I've developed my.

38:14 Oil land into plantations.

38:17 Now I'm attracting a local labor market,

38:18 for example,

38:19 and that's making it cheaper for you

38:21 to produce.

38:22 That's a very realistic,

38:24 natural spatial interaction that I'm not going to be able to accommodate.

38:27 Why?

38:28 Because it's going to end up being hard to relax,

38:30 one,

38:31 absent a model of local labor markets and absent data on local labor markets,

38:35 which I don't have.

38:36 But two,

38:37 It's going to be hard to relax because if I'm a big player and I delay,

38:40 then others are going to respond

38:42 to that delay,

38:43 that I,

38:44 my action,

38:45 and that's going to change how the

38:47 economy is evolving and where the state of the economy is going.

38:50 That is to say,

38:51 now the continuation values are no longer going to align

38:54 and difference out,

38:55 and then we're going to have computational issues.

38:57 Now,

38:57 lastly,

38:58 I'm going to be assuming rational expectations,

39:00 which by the way,

39:01 does mean that expectations can be wrong today.

39:04 They just need to be correct over time.

39:06 Of course,

39:07 still a very strong assumption,

39:08 but it's going to be much weaker than needing

39:10 to specify what exactly expectations are in every period,

39:14 like we would do

39:15 with a conventional full solution approach.

39:19 Now,

39:19 the upshot of this is we're going to be

39:20 able to estimate the model with a linear regression,

39:23 and so I can talk about endogeneity in instruments in familiar ways,

39:27 and furthermore,

39:28 Euler methods are going to accommodate the non-stationarity of the problem.

39:31 So I actually can't use typical CCP methods like BBL or POB

39:36 because those require stationarity.

39:38 And

39:39 if,

39:39 you know,

39:39 BBL and POB are just letters to you,

39:41 then

39:41 you can ignore that sentence,

39:42 not a,

39:42 not a,

39:43 not central.

39:44 OK.

39:45 So let's start with uh estimation,

39:47 uh,

39:48 what,

39:48 what I'm estimating,

39:49 and let's start with the continuous choice on the intensive margin.

39:53 Now,

39:53 uh,

39:54 if

39:55 this is the plantation development decision,

39:57 that continuous twist.

39:59 This slide I want to flag is pretty standard.

40:01 It's just like the consumption,

40:02 consumption savings problem in,

40:04 you know,

40:04 grad school macro 10 1.

40:06 So I'm gonna go through quickly,

40:07 but I,

40:08 I,

40:08 I'm still gonna go through because the intuition is useful.

40:11 So,

40:11 in this top line,

40:12 the Euler equation is capturing that trade-off between investing today

40:16 or delaying.

40:17 Now,

40:17 this is a first-order condition.

40:19 If I invest today,

40:20 then revenues are higher because I'm going to start producing earlier,

40:23 but at the same time,

40:24 costs are also higher because I have to pay them today

40:27 instead of later when they're discounted.

40:30 So this is going to specialize to this second line.

40:32 I'm going to develop more plantations today,

40:34 AIT,

40:36 uh,

40:36 relative to tomorrow,

40:37 AIT plus 1,

40:38 for action,

40:39 if the extra revenues are high or the extra costs are low.

40:43 Now,

40:44 note that that top line has expectations,

40:46 but they don't show up in the

40:48 second line because I'm substituting them out

40:50 for realized values

40:52 plus expectational errors.

40:54 This is the usual trick.

40:56 Uh,

40:56 now,

40:56 under rational expectations,

40:58 those errors are mean zero conditional on anything in the period T

41:03 information set.

41:03 Note,

41:04 of course,

41:04 that we're talking here about

41:05 uh RT + 1

41:07 and not RT,

41:09 OK?

41:09 And so we need to be a little bit careful.

41:10 RIT + 1,

41:11 not RIT.

41:12 So we need to be a little bit careful here.

41:14 I need to use the lags.

41:16 So,

41:16 let's then talk about this linear regression that we get,

41:20 uh,

41:20 and let's talk about identification.

41:22 First off though,

41:22 what are the data

41:24 here?

41:25 Uh,

41:25 the left-hand side is data.

41:26 OK,

41:27 so I see in the satellite data how much plantation development A

41:31 or action there is in each site I and in each period.

41:35 AIT AIT plus 1,

41:36 I see it.

41:37 Beta discount factor,

41:38 generically unidentified,

41:40 you know,

41:40 Manac Desmar,

41:40 uh,

41:42 you know.

41:43 Not identified,

41:44 I'm just gonna set it.

41:45 I can play around with it though and show you robustness.

41:48 OK,

41:48 so the left-hand side,

41:49 I see everything.

41:50 And then

41:51 prices P

41:52 yields Y and cost factors like road distance,

41:55 urban distance X.

41:57 I see all of those,

41:58 those are data.

41:59 So then what am I actually estimating here?

42:02 The goal is to estimate these parameters data of the payoff function.

42:07 And then the error term is going to have

42:09 two kinds of errors that we need to worry about.

42:11 One,

42:12 the unobserved cost shocks epsilon,

42:14 as well as two,

42:15 those expectational errors.

42:17 Actually,

42:18 both are going to cause us problems.

42:19 So let's start with the epsilons.

42:21 Those are going to cause price endogeneity for the following reason.

42:24 Low costs today

42:26 mean high entry today,

42:27 so high supply tomorrow,

42:29 and therefore low prices tomorrow.

42:32 Now,

42:32 the atas are also correlated for the reason I said before.

42:34 This is PT plus 1,

42:36 not PT.

42:37 So for the instruments,

42:38 I need to use lags.

42:40 Um,

42:40 OK,

42:41 so price endogeneity on the supply side,

42:43 what do we want?

42:43 We want a demand shifter.

42:45 So I have that demand shifter.

42:46 I'm going to get that demand shifter

42:48 from

42:49 demand estimation,

42:50 that's that right word shift over time in the

42:52 demand curve that I plotted for you before.

42:54 Of course,

42:54 that's just time series variation at the end of the day.

42:56 OK,

42:57 so here what it's going to do is it's going to interact

42:59 with that cross sectional variation in yields,

43:02 which I can instrument for with exogenous factors like sunlight.

43:06 So here's the intuition.

43:07 When prices go up,

43:08 it's going to increase revenues for sites that can actually grow palm oil.

43:12 It's not going to help very much for a site that just can't grow palm oil to begin with.

43:16 Why?

43:16 Because,

43:17 you know,

43:17 revenue is price times quantity,

43:19 so if quantity is zero,

43:20 then revenue is always zero.

43:22 Price going up doesn't matter.

43:23 So that is to say that suitable sites

43:25 are going to be more likely than unsuitable sites

43:27 to develop land when prices rise,

43:30 and that difference is going to give me the supply elasticity

43:34 that I'm after.

43:35 And

43:36 then there's then there's a question about.

43:39 I'm good.

43:41 So yeah,

43:41 I have two questions.

43:42 Uh,

43:43 first is,

43:44 can,

43:44 can you store palm oil at all?

43:46 Uh,

43:47 so,

43:47 uh,

43:48 palm oil can be stored.

43:49 I don't have that either on the demand side or the supply side.

43:52 On the demand side,

43:53 side,

43:53 I actually see the storage of palm oil.

43:55 You might be worried about a stockpiling

43:58 concern and dynamic demand.

43:59 It turns out to be very small,

44:01 and,

44:01 and I,

44:01 I see that it's small because it's in the data.

44:03 On the supply side,

44:05 wait one second.

44:06 What does it mean it's small

44:08 storage is,

44:08 is,

44:09 is non-zero,

44:09 then,

44:10 then you have.

44:11 Then you have the hotening conditions kicking in.

44:14 Yeah,

44:14 but the thing is,

44:15 if you look at the kind of ketchup papers or the laundry detergent papers,

44:18 so Handel Nouveau,

44:19 but also

44:20 Michael Keene has work on this,

44:22 there you can see they kind of back out,

44:24 OK,

44:24 for a given amount of consumption.

44:27 Every day I'm consuming this much relative to consumption.

44:30 How much do I have stored at any given period of time?

44:32 And you see,

44:33 I mean,

44:33 we see this with olive oil that we just use at home,

44:35 right?

44:36 How much do we use today relative to how much is in the bottle?

44:38 We're storing,

44:39 you know,

44:40 3,000% of what we're consuming day to day.

44:43 Here,

44:43 because it's part,

44:44 and part of this is just because of the aggregation,

44:46 it's at the country level and it's year on year.

44:49 You see that of the amount that is consumed every year,

44:51 it's only like 10%,

44:53 it's something like 10%

44:55 being stockpiled at any given time.

44:57 So this is the sense in which,

44:58 you know,

44:59 that stockpiling can go from

45:00 10% to 15%,

45:02 but it's not going to buy us things in a way.

45:04 I understand,

45:04 but the fact that

45:05 that the stockpile is non-zero and and the fact it's only 10% can be the outcome of.

45:11 Of an economy which is driven by,

45:13 by,

45:15 by prices

45:16 in the hoteling world.

45:17 I,

45:18 I,

45:18 you know.

45:20 The issue is,

45:20 is as soon as there is a potential for speculative storage,

45:24 then your price equation becomes pinned down by,

45:27 by,

45:28 by a different equation.

45:29 Yeah,

45:30 I agree.

45:31 No,

45:31 I agree.

45:31 This is going to be a fundamental issue with

45:33 at the end of the day,

45:34 my demand is static demand,

45:36 not dynamic demand.

45:38 We can talk about making a dynamic demand.

45:40 I'm not sure at the end of the day it's going to give me some price,

45:43 some price elasticity,

45:44 demand elasticity that's going to speak to the leakage question.

45:47 We can talk about if dynamic demand would change that price.

45:50 Elasticity

45:51 from 0.7 to 7.

45:52 If that's the case,

45:53 OK,

45:53 it's super important.

45:55 Uh,

45:55 but if it's going to go from 0.7 to 0.8,

45:57 then it's not going to change qualitative conclusions by that much.

46:00 But yes,

46:00 I totally agree with this.

46:02 No,

46:02 I'm talking on the,

46:03 I'm talking on the supply side.

46:04 I'm talking not talking about demand in Europe.

46:06 I'm talking about Indonesia traders or the stockpiling in order to,

46:10 to sell strategically

46:12 given

46:12 the crisis that are going to happen.

46:14 I think that's the most,

46:16 most problematic problem because this is how you get your amplification.

46:21 So your PT +1,

46:22 YT plus one.

46:24 Fully hedges on the fact that there is no,

46:26 that,

46:27 that your price opportunity is from a market clearing condition,

46:29 not on the storage condition.

46:31 Yes,

46:31 yes,

46:32 so I,

46:32 I agree with that.

46:33 That,

46:33 that's,

46:34 that on the supply side,

46:35 I,

46:35 I am missing that storage margin.

46:38 Here again,

46:38 so I,

46:39 I,

46:39 I

46:40 don't have the,

46:40 the data off the top of my head,

46:42 but also this stockpiling

46:44 within Indonesia is also observed.

46:45 And so one track that I haven't done but should do,

46:48 and,

46:48 and thank you for Bringing it up is

46:49 to see the extent to which that stockpiling is responding to

46:53 the historical price variation,

46:54 and I can

46:55 start to at least speak to this more concretely in that way.

46:58 Yeah,

46:59 here,

46:59 when you produce it,

47:00 you have to sell it right away

47:01 and,

47:01 and therefore,

47:02 yes,

47:02 the,

47:02 the price variation is just coming from,

47:04 from,

47:05 uh,

47:06 you know,

47:06 the,

47:07 the,

47:07 the kind of price variation

47:08 period to period.

47:10 Um,

47:10 in counterfactuals,

47:11 so that's an estimation concern.

47:13 For counterfactuals,

47:13 I'm not as worried,

47:14 you know,

47:15 conditional on the estimation being right,

47:17 because here you might think that,

47:18 you know,

47:19 I try to tie,

47:20 I see a tariff coming into play,

47:21 and so I try to rush by selling all of my stockpile right now,

47:24 for example,

47:25 um,

47:25 or I try to rush into

47:27 putting,

47:28 setting up a farm to kind of time out the tariffs being put in place.

47:31 But here,

47:32 tariffs are being put,

47:33 are,

47:33 are,

47:33 are in place on consumption in every future period.

47:36 OK,

47:36 so there's kind of a limit to the extent.

47:38 To which you can kind of rush

47:40 and uh

47:41 preempt the tariff in the way that you can in,

47:43 in other environmental

47:45 uh settings.

47:47 No,

47:47 no,

47:47 I'm fine,

47:47 but

47:48 my only concern is,

47:49 you know,

47:50 when,

47:50 when you showed your,

47:51 your,

47:51 your price trend,

47:52 your log price that is very stable over time,

47:55 it does look like

47:57 a storage,

47:58 uh,

47:59 price pattern.

48:00 Yes,

48:01 I mean it's very constant returns.

48:02 So,

48:02 so that's why I,

48:03 I,

48:04 no,

48:04 no,

48:04 totally you make your reason not to have it,

48:06 but that's what I wanted to try.

48:07 That's it.

48:07 Yeah,

48:08 absolutely.

48:08 No,

48:08 I,

48:09 I agree with that.

48:09 Although the one kind of

48:11 other point here is that if demand is just going up over time,

48:15 as it is,

48:15 you know,

48:16 very quickly,

48:17 then actually if you,

48:18 if you

48:19 expected this storage component to be really,

48:21 really meaningful,

48:23 and if palm oil can be stored for 567 years with no problem,

48:26 as it can be.

48:27 Then you would expect that everyone would just stockpile everything all

48:30 of the time and never sell anything given that no,

48:32 no,

48:32 no,

48:32 no,

48:32 no,

48:32 no,

48:32 no,

48:33 that's what the hotel and coalition tells you is people

48:35 in every period are indifferent between selling and stockpiling.

48:38 So it doesn't have any prediction on the quantity of stocks,

48:41 but this pins down

48:42 instead of having market con what I'm,

48:44 my point is instead of having the market clearance condition to pin down your price

48:48 in,

48:49 in your early equation,

48:50 it's a hotening condition that pins down the price.

48:52 I see,

48:53 I see that's a completely different,

48:55 that's,

48:55 that's a completely different early equation,

48:57 yeah,

48:57 because it's,

48:57 it's,

48:57 it's off of the storage versus selling as

48:59 opposed to selling today versus selling tomorrow.

49:01 Yeah,

49:02 producing today and producing tomorrow.

49:04 Yeah,

49:04 I,

49:04 I've coupled producing and selling.

49:06 Your point is that

49:07 producing and selling are actually two,

49:10 when we decouple it,

49:11 it becomes

49:12 more complicated.

49:12 You have this extra condition.

49:14 Yeah,

49:14 I,

49:14 I agree with that.

49:15 OK,

49:15 great.

49:16 So let's try to

49:17 move on to the results,

49:19 uh,

49:19 in the interest of time because we have about 20 minutes left.

49:22 Great.

49:23 Um,

49:23 OK,

49:24 so let me talk to you about the extensive margin.

49:27 So here,

49:27 what's happening,

49:28 uh,

49:28 is that we have this,

49:30 uh,

49:30 mill being built.

49:32 OK,

49:32 so this is that discrete choice.

49:34 Uh,

49:34 so if

49:35 this were the only margin,

49:36 then we could just use the usual Euler CCP

49:39 methods directly.

49:40 So,

49:40 let me Give you that intuition first.

49:42 Either we can invest today or delay.

49:44 Either way,

49:44 the mill is being built by T + 2,

49:46 so the

49:48 continuation values align,

49:49 we have finite dependence.

49:50 Now,

49:50 I am imposing that the mill actually gets built in that delayed case.

49:54 That's potentially sub-optimal.

49:55 It's going to depend on the loaded shock.

49:57 So this correction term gamma is going to account for that.

49:59 Now my model is a little different because

50:01 we also have this continuous intensive margin.

50:03 The problem here is going to be that delaying the mill also delays the plantation.

50:07 If I delay and only plant AIT,

50:09 then the continuation values don't align.

50:11 And so instead,

50:12 we need to catch up to that original path

50:15 to get finite dependence.

50:16 Again,

50:17 I'm imposing a particular potentially suboptimal choice.

50:19 The correction term needs to be expanded to account for that.

50:22 So combining this continuous,

50:24 these continuous and discrete pieces is the

50:27 marginal methodological contribution here.

50:29 This is common,

50:30 a common setup.

50:31 Businesses make a discrete choice to enter

50:33 a market and then a continuous choice over how much to produce,

50:36 but I don't want to overclaim here because

50:38 CCP methods could already deal with both margins.

50:40 It's just

50:41 oiler CCP methods

50:42 that focused on the extensive margin,

50:44 and I'm relaxing that.

50:46 So

50:46 from here we're going to get a regression.

50:48 This is a two-step estimator,

50:49 just like the usual CCP methods.

50:52 Without going into detail,

50:53 what I'll just say

50:55 is that

50:55 this piece here is where the two margins hook together.

50:58 OK,

50:58 so this is where the intensive margin

51:00 estimates

51:01 come in.

51:02 And what's the intuition here?

51:03 It's just that I'm more likely to build

51:05 the mill in the first place if those plantations

51:08 would actually be good.

51:09 If the plantations will be terrible,

51:10 then why build the mill to begin with?

51:13 OK,

51:13 so now I'm going to have the cost structure of the

51:15 industry which I'm going to need of course for counterfactuals.

51:17 Here are those cost estimates,

51:19 estimates.

51:20 So

51:20 I find that on average to develop one hectare of land into plantations,

51:24 I get an average cost of

51:26 $10,000.

51:27 That's going to line up.

51:28 With the $7000 accounting cost,

51:31 different

51:31 economic objects,

51:32 of course,

51:32 but we're in the same ballpark,

51:34 so this is at least one way of

51:35 starting to validate

51:37 the estimates.

51:38 Here cost factors don't enter because at this point the mill has sunk.

51:40 You just kind of build the plantations around the mill.

51:44 So for mills,

51:45 I'm going to get a $23 million average

51:46 cost compared to $20 million in accounting costs,

51:49 and here the cost factors do matter.

51:51 So being close to major roads,

51:52 major ports,

51:53 major urban centers,

51:54 major urban centers is going to be less costly because you have

51:58 lower transportation costs.

52:00 Um,

52:00 now we see here that sites actually do try to avoid forest.

52:03 That's good.

52:04 Maybe it's just because then they need to pay a bribe if they need to.

52:06 Go and cut down all of these trees,

52:08 but we see here that there's no accounting at all for peat.

52:11 OK,

52:11 so cutting,

52:12 releasing peat emissions is completely

52:15 inconsequential and not costly at all to these firms in terms of private cost,

52:19 and that's where all of the big emissions are.

52:21 OK.

52:21 And so this is the sense in which private costs

52:23 are failing to line up with social costs,

52:25 and this is the externality that we want to force firms to internalize.

52:30 OK,

52:31 so let me put the pieces together for you now and we can talk about

52:33 counterfactuals.

52:34 So,

52:35 suppose the EU puts in tariffs tar,

52:37 conditional on supply,

52:38 sellers are now going to be profiting less from the EU

52:41 and so they're going to shift sales over to China.

52:43 Now the demand elasticities are going to determine how

52:46 much shifts over to China and therefore how much

52:49 world prices fall.

52:50 and lower world prices are going to translate into less supply given the supply.

52:55 Now I have data on the carbon stocks,

52:56 where the trees and the peat

52:58 are,

52:58 and so

52:59 when supply contracts in this region versus that region,

53:02 I'm going to be able to compute the

53:05 implications for carbon reductions directly from the data.

53:09 Now I am going to monetize

53:10 emission reductions with the $40 social cost of carbon.

53:14 And then,

53:14 and this is getting at uh the question from before,

53:16 so I'll not rehash everything.

53:18 I am assuming that

53:19 palm oil tariffs are not increasing

53:21 non-palm deforestation.

53:23 Acacia is the big threat there,

53:24 but I check

53:25 uh that substitution in the data,

53:27 and it's,

53:27 it's small anyway,

53:28 so I didn't augment the model to

53:30 account for that.

53:31 Uh,

53:31 otherwise,

53:31 it would have been important to,

53:33 to account for it directly.

53:34 Now,

53:36 the

53:37 other thing that I'll just point out here is that,

53:40 you know,

53:40 sometimes this gets a little lost.

53:41 This is an equilibrium model.

53:43 You know,

53:43 there's this leakage across markets.

53:45 Prices are set in equilibrium,

53:47 farmers are playing a dynamic competitive equilibrium,

53:50 but this is only within palm oil.

53:52 OK,

53:52 so I don't have effects via other industries

53:54 because that would require a multi-industry model.

53:57 And so these trade computable general equilibrium models are

54:00 going to do much better at capturing that cross-industry

54:03 kind of leakage and those cross-industry effects,

54:05 but the cost is going to be that

54:07 the

54:08 production functions within each industry and those models is going to be much

54:11 more simple and a little bit crude compared to what I have,

54:14 where I can really give you good detail within this one industry.

54:18 OK,

54:19 so in the baseline,

54:19 I'm gonna set tariffs to maximize global welfare.

54:22 I can relax that,

54:23 and,

54:23 and in that case,

54:24 the terms of trade effects

54:26 or the kind of

54:27 Indonesian market power effects would come in.

54:29 Uh,

54:29 I do do that in,

54:30 in robustness.

54:32 Here I'm also to get at Ariana's question,

54:34 I am

54:35 treating all palm oil equally,

54:36 again because it's much easier to administer,

54:38 but I can and do relax that assumption as well.

54:41 I can also differentiate between good palm oil and bad palm oil.

54:45 So for leakage,

54:46 I'm going to show that the

54:49 effects of different,

54:50 I'm going to show you the effects of different coalitions.

54:52 So

54:52 all importers,

54:54 that's the big coalition or just the EU alone,

54:56 that's the small coalition.

54:58 I'm not going to be modeling the coalition formation game,

55:00 but I am going to be able to quantify for you

55:02 the incentives

55:03 to defect.

55:04 Uh,

55:05 now for commitment,

55:07 full commitment is tariffs upheld forever.

55:09 There's no game here because the regulator moves once at the beginning

55:12 and then producers follow.

55:13 There's no back and forth.

55:15 Uh,

55:15 no commitment is just sequential static optimization.

55:18 And then lastly,

55:19 limited commitment is like a Chinese 5-year plan.

55:22 OK,

55:22 you set it for 5 years,

55:23 you do it,

55:24 and then you reset again,

55:26 uh,

55:26 after 5 years are over.

55:28 And so,

55:28 uh,

55:28 a 2-year

55:29 plan is very close to sequential static

55:31 optimization because you're updating a lot.

55:33 A 50-year plan is like

55:35 full commitment because you don't update that much at all.

55:38 So note here there is going to be this back and forth,

55:41 there is a game

55:42 because

55:42 the regulator moves and then the producers

55:44 respond and then the regulator moves again,

55:46 taking producer's past response as given.

55:49 So I'm going to need to solve by backward induction.

55:52 Um,

55:53 note here,

55:53 and this sort of came up a bit before,

55:55 but I'll just,

55:55 ah,

55:56 bring it up again to clarify,

55:58 producers are atomistic and competitive,

56:00 so there's no sense in which they're gonna

56:02 rush to clear land and push tariffs to zero,

56:04 because if you can cut down all of the land,

56:06 then tariffs

56:07 should go to zero optimally.

56:08 But they can't do that because they're,

56:10 they're atomistic.

56:11 Uh,

56:11 they also can't really rush to get in before the tariffs,

56:14 and that's,

56:15 and that's the green paradox because tariffs

56:17 are getting paid in every future period.

56:18 So you can rush and avoid tariffs for one period,

56:20 but you're still paying it

56:22 over the future horizon.

56:24 Now,

56:25 for the sake of time,

56:26 I'm gonna skip how I,

56:28 the technical details of how I actually solved the model.

56:31 Let me just say this one point.

56:33 So,

56:33 when I estimated the model,

56:36 I estimated it without actually solving it,

56:38 because the Euler methods allowed me to do so.

56:40 That meant two things.

56:42 Excuse me.

56:43 Number 1,

56:44 computation was much easier,

56:45 and number 2,

56:46 there were fewer assumptions that I needed to make

56:48 on expectations.

56:50 Here,

56:50 in counterfactuals,

56:51 I actually need to solve the model.

56:52 There's no getting around it.

56:54 And so that,

56:55 again,

56:55 means two things.

56:56 Number 1,

56:56 computation is gonna be harder.

56:58 I'm gonna need to use some computational tricks here.

57:00 And 2,

57:00 I am gonna need some more assumptions on expectations.

57:03 I'm gonna need to actually specify them.

57:05 Uh,

57:05 but we can talk about that in if there's time for questions.

57:09 So let me just get into the,

57:10 the counterfactuals,

57:11 the counterfactual results.

57:13 So the baseline result here is that coordinated committed tariffs are effective,

57:17 OK?

57:17 How do we see that here?

57:18 The Y axis is emission reductions,

57:20 the higher is better,

57:22 and the x axis is the commitment period.

57:25 The solid blue line is coordinated tariffs.

57:28 So committed coordinated tariffs are right over here,

57:30 and we see they're pretty close to that,

57:32 uh,

57:33 domestic,

57:34 socially optimal domestic tax,

57:35 which I'm thinking of as infeasible,

57:37 but will be our,

57:38 our benchmark.

57:39 So there are two caveats here.

57:41 Number 1,

57:41 there are going to be pretty big losses as coordination breaks down,

57:44 down to the EU alone.

57:46 So for the EU alone,

57:46 for example,

57:47 leakage concerns are going to kick in because the EU

57:50 is acting alone and because it has low coverage,

57:53 and that those leakage concerns are going to result in small tariffs.

57:56 So losses here are disproportionate because low coverage

57:59 is going to get compounded by low tariffs.

58:02 Now,

58:03 0.2 is that there are big losses as commitment breaks down,

58:07 and the starkest case here is no commitment,

58:09 where tariffs never get applied because of time to build.

58:12 So,

58:13 the intuition here is that tariffs today are going

58:14 to get set to zero because they can't produce,

58:17 uh,

58:17 they can't prevent new plantations,

58:19 because there's no new plantations aren't producing anything that's taxable

58:22 until tomorrow.

58:23 So those new plantations are really only thinking about tariffs tomorrow,

58:26 not tariffs today.

58:28 But then by the time we get to tomorrow,

58:30 those new plantations aren't going to,

58:32 uh,

58:34 by the time we get to tomorrow,

58:35 tomorrow's tariffs are again going to be set to zero because

58:38 from the perspective of tomorrow,

58:40 the plantations that were new today are already sunk.

58:43 OK,

58:43 and so tariffs tomorrow can't change

58:45 the fact that those plantations have been built,

58:47 and so again,

58:48 tariffs are going to be set to zero,

58:49 that is to say they're always going to be zero.

58:52 But there's a

58:52 positive side here.

58:54 So number 3,

58:55 all of these policies are pretty efficient relative to other

58:58 things like retrofitting power plants for carbon capture and storage.

59:02 So unilateral EU

59:03 action,

59:04 yes,

59:04 has a pretty small effect,

59:06 2 or 3% compared to,

59:07 you know,

59:07 65%.

59:08 percent

59:09 in the,

59:09 in that domestic intervention benchmark.

59:12 So 2 to 3% is small,

59:13 but first off,

59:14 2 to 3% is better than nothing.

59:16 And then number 2,

59:16 at that small scale,

59:18 it's only $10 of

59:20 surplus for gun

59:21 per ton of carbon averted.

59:23 OK,

59:23 so pretty efficient.

59:24 And then number 4,

59:25 coordination is going to help us alleviate also the commitment problem.

59:29 So let me rescale those blue lines for you.

59:32 Under full coordination,

59:34 that's that solid blue line,

59:35 we see that committing for 5 years at a time is actually enough to achieve 95%

59:40 of the full commitment outcome.

59:42 And 5 years is salient because that's the political cycle.

59:45 On the other hand,

59:45 if the EU is acting alone,

59:47 then it needs to commit for 20 years at a time to get to

59:50 that 95% efficiency relative to the,

59:53 the full commitment outcome.

59:55 OK,

59:55 and so this is the sense in which leakage and commitment interact.

59:59 That is,

59:59 as you make progress on coordination,

1:00:01 you're alleviating the leakage problem and

1:00:03 you're also alleviating this commitment problem.

1:00:07 Now I can also zoom in on the division of surplus across markets,

1:00:10 and here that exercise is going to show us that actually,

1:00:13 unfortunately,

1:00:14 coordination and commitment are both difficult in practice.

1:00:18 So first off,

1:00:18 there's going to be this incentive

1:00:20 not to coordinate.

1:00:22 So suppose the EU,

1:00:23 China and India are all acting and all importers are and other importers here,

1:00:27 Nigeria,

1:00:28 Bangladesh,

1:00:28 the US,

1:00:29 etc.

1:00:29 are thinking about joining the coalition.

1:00:32 So if they do,

1:00:33 then they're going to sacrifice,

1:00:34 they're going to be sacrificing consumer surplus.

1:00:36 We see consumer surplus goes down.

1:00:38 They're happy to do so because it reduces emissions if they care about emissions.

1:00:42 If they don't care about emissions and only care about consumer surplus,

1:00:45 there's no point in sacrificing their consumer surplus

1:00:47 and therefore no point in imposing import tariffs themselves.

1:00:51 And furthermore,

1:00:51 here they're going to be able to free ride

1:00:53 because the tariff coalition in acting,

1:00:56 that is the EU,

1:00:57 China,

1:00:57 and India in acting,

1:00:58 are putting in tariffs and pushing down world prices.

1:01:01 And so now these other importers compared to other

1:01:04 worlds where no one is doing anything are actually even better off

1:01:07 because they're enjoying lower,

1:01:08 lower world prices.

1:01:11 Uh,

1:01:11 so this gap then is the transfer that we need to think about

1:01:14 in order to induce participation,

1:01:16 uh,

1:01:17 into the coalition by these actors.

1:01:20 So there's going to be a very similar incentive not to commit,

1:01:23 where,

1:01:24 you know,

1:01:24 if you commit for a long period of time,

1:01:26 that's increasing tariffs,

1:01:27 and the point of doing that is

1:01:28 more consumer surplus sacrifice

1:01:30 for bigger

1:01:31 carbon reductions.

1:01:32 But again,

1:01:32 that only makes sense if countries care about carbon reductions

1:01:36 and carbon emissions.

1:01:38 However,

1:01:38 note that under coordination,

1:01:39 actually this gap here is not really very big,

1:01:42 um,

1:01:43 but when coordination breaks down,

1:01:44 so here the other importers have left,

1:01:46 it's just the EU,

1:01:47 China and India left,

1:01:48 that gap we see is,

1:01:49 is bigger.

1:01:50 So again,

1:01:50 this interaction between

1:01:52 uh coordination

1:01:53 and commitment.

1:01:55 And then last,

1:01:56 I can show you those effects on Indonesian and Malaysian surplus.

1:02:01 So if the problems are only weak incentives for domestic regulation,

1:02:04 then actually here tariffs can help

1:02:06 a bit.

1:02:07 So suppose importers threaten Indonesia and Malaysia.

1:02:10 That is to say,

1:02:10 either you put in the large socially optimal domestic tax

1:02:14 or face tariffs.

1:02:16 What we see here is that in most cases,

1:02:18 you know,

1:02:19 Indonesia and Malaysia are better off just accepting those tariffs,

1:02:21 which are smaller and therefore hurt producer surplus

1:02:24 by less.

1:02:25 Uh,

1:02:25 they'll just accept them.

1:02:26 But if the threat on the other hand,

1:02:28 is coordinated tariffs,

1:02:29 then we see that actually Indonesia and Malaysia are better off just moving to

1:02:33 that socially optimal domestic tax,

1:02:35 which one isn't very much bigger than coordinated tariffs anyway.

1:02:39 Coordinated tariffs are already pretty big.

1:02:41 That is to say,

1:02:41 in moving to this domestic tax,

1:02:43 producer surplus is not going down by very much.

1:02:46 However,

1:02:46 now,

1:02:47 Indonesia and Malaysia are able to get that government revenue

1:02:49 from that tax that was before going abroad to Europe,

1:02:53 you know,

1:02:53 China,

1:02:54 India,

1:02:54 etc.

1:02:56 Uh,

1:02:56 now,

1:02:57 we might also think,

1:02:59 uh,

1:02:59 and this came up before,

1:03:00 that

1:03:01 an export tax is a smaller ask that's maybe easier to administer.

1:03:05 So administratively it's easier,

1:03:07 as I said before,

1:03:08 because now we only need 5 auditors at the ports

1:03:10 instead of 1000 auditors to cover all the forests.

1:03:13 Uh,

1:03:13 but you know,

1:03:14 again,

1:03:14 the counterpoint is that maybe companies now only need

1:03:16 to bribe 5 people instead of 1000 people,

1:03:18 maybe it's easier.

1:03:19 Um,

1:03:20 but either way,

1:03:21 uh,

1:03:21 we,

1:03:22 if we can put in the socially optimal export tax,

1:03:25 actually we're going to achieve exactly the

1:03:27 same emission reductions as these coordinated tariffs,

1:03:30 because in both cases we're covering all imports.

1:03:32 That is to say we're covering all exports out of Indonesia and Malaysia.

1:03:36 Uh,

1:03:36 but note,

1:03:36 however,

1:03:37 that

1:03:37 these

1:03:38 uncoordinated tariffs over here

1:03:40 are

1:03:41 also still not enough to push Indonesia and Malaysia to this export tax.

1:03:45 Uh,

1:03:46 and then also that export tax might increase,

1:03:48 uh,

1:03:48 as Orellana had brought up before,

1:03:50 this palm oil production outside of Indonesia and Malaysia,

1:03:53 in West Africa and South America,

1:03:55 which import tariffs would

1:03:57 kind of catch by default.

1:03:59 Uh,

1:03:59 so,

1:04:00 lastly,

1:04:00 note here,

1:04:01 and this gets to Tristan's equity concern from the very beginning.

1:04:05 OK,

1:04:05 note that all regulation hurts Indonesia

1:04:08 and Malaysia.

1:04:08 Everything is negative across all of these policies.

1:04:11 OK,

1:04:11 and so,

1:04:12 uh,

1:04:12 if we

1:04:13 have these equity concerns,

1:04:14 which we should,

1:04:15 of course,

1:04:16 then

1:04:16 at least one,

1:04:17 I'm going to be able to quantify those damages for you,

1:04:21 and therefore,

1:04:21 2,

1:04:22 we can talk about the transfers that might be needed to address

1:04:25 these equity concerns as,

1:04:26 as much as we can.

1:04:29 Uh,

1:04:29 so,

1:04:30 let me just take 30 seconds to conclude,

1:04:32 and then we'll have some time for questions.

1:04:34 Um,

1:04:35 OK,

1:04:35 so to summarize what we talked about today,

1:04:37 we've talked about how import tariffs can be effective if they're coordinated

1:04:41 and if they're committed.

1:04:42 And this is going to help when domestic regulation is infeasible,

1:04:45 including when those domestic issues maybe take time to change,

1:04:49 but we want to act on the climate today

1:04:51 and now.

1:04:52 And then as we're setting,

1:04:53 we look at the palm oil industry,

1:04:55 which accounts for a massive 5%

1:04:57 of global carbon emissions.

1:04:59 But I want to end on a,

1:05:01 on a positive note.

1:05:02 Even though that past deforestation is sunk and it's over,

1:05:06 there are still huge swathes of forests that remain intact,

1:05:09 especially in Papua and also to a lesser extent on Borneo.

1:05:13 And that's the scope of what we can still save if

1:05:16 we go ahead and act and implement good policy today.

1:05:19 Thank you.

1:05:22 OK,

1:05:23 that's great.

1:05:23 Thanks for sticking to the time,

1:05:25 Govinda.

1:05:26 Sorry,

1:05:26 I had another question.

1:05:28 Alan,

1:05:28 I'm just wondering about your CO2 reduction coefficient.

1:05:32 Uh,

1:05:33 the,

1:05:33 if the palm oil is used for biodiesel,

1:05:36 so in that case,

1:05:37 the biodiesel substitute the fossil fuel and

1:05:40 reduce the emission on the demand side.

1:05:42 So whether you have accounted for that reduction

1:05:45 in the emission in the supply side.

1:05:48 Yes,

1:05:48 so actually,

1:05:49 I,

1:05:49 I did that calculation.

1:05:50 When you have biofuels,

1:05:51 you're averting kind of fossil fuel emissions.

1:05:54 It turns out that those fossil fuel emissions averted

1:05:57 are very small compared to the kind of emissions needed to,

1:06:00 to grow palm oil,

1:06:02 uh,

1:06:02 on average.

1:06:02 Um,

1:06:03 those palm oil emissions are so big that

1:06:05 even if you account for that,

1:06:06 it's,

1:06:06 it's still,

1:06:07 you know,

1:06:07 it's on the order of 5%.

1:06:09 Yes,

1:06:09 you avert a little bit,

1:06:10 but not nearly enough to account for the peat that you've destroyed.

1:06:13 That's why Europe has removed,

1:06:15 they used to subsidize palm oil

1:06:17 for use in biofuels,

1:06:18 and,

1:06:18 and

1:06:19 just in the past couple of years,

1:06:20 they've removed those subsidies.

1:06:22 Uh,

1:06:22 and so it's just a little bit after my,

1:06:23 my sample,

1:06:24 but,

1:06:24 um,

1:06:25 they removed those subsidies because they realized,

1:06:27 OK,

1:06:27 actually,

1:06:27 we really should not be subsidizing biofuel,

1:06:30 uh,

1:06:30 palm oil for use in biofuels.

1:06:32 Not only that,

1:06:33 the EU already banned use of palm oil for the biodiesel production starting in 2020.

1:06:41 Yes,

1:06:41 although it's,

1:06:41 it's not a full-on ban,

1:06:42 it's,

1:06:43 it's phased out over time.

1:06:44 Originally it was a full-on ban,

1:06:46 and then,

1:06:46 you know,

1:06:46 Indonesia said we're going to cancel all of our Airbus,

1:06:49 purchases,

1:06:50 for example.

1:06:51 And

1:06:51 so there's diplomatic,

1:06:52 there's a bit of a diplomatic row,

1:06:54 and there was a WTO lawsuit,

1:06:56 and so they shifted it to,

1:06:57 I think it's by 2030 now.

1:06:58 Everything's going to get phased out.

1:06:59 Norway though,

1:07:00 however,

1:07:00 Norway's not an EU country,

1:07:01 of course,

1:07:02 but closely related.

1:07:03 Uh,

1:07:03 Norway just full-on banned and,

1:07:04 and,

1:07:05 and called it,

1:07:06 called it a day.

1:07:12 OK,

1:07:12 that's great.

1:07:13 If there are no further questions,

1:07:14 then let's conclude the seminar and,

1:07:16 uh,

1:07:17 thank you very much,

1:07:18 Alan.

1:07:18 It was very interesting.

1:07:20 Uh,

1:07:20 thank you for meeting both.

showAllTimestamps
no
transcript
for palm oil and in substitutes. Now that's that first issue. Second, there's going to be this commitment problem, and here's where the dynamics kick in. OK, so once those forests are cut down, ex post, there will now be this temptation to reduce tariffs. Why? Because after the forest is gone, those emissions are sunk. But on the supply side, farmers are thinking about their future palm oil revenues. Again, so then the problem is this. Without commitment, future tariffs are low. And so if I'm a farmer, I'm just going to keep on deforesting today, even if today's tariff happens to be high. Why? Because I'm thinking about the future mostly. So, the dynamics are really going to matter economically here. Uh, they're also going to turn out to matter quantitatively. But dynamics, of course, are going to make estimation much harder computationally. So I'm also going to make a small methodological contribution here, building on existing oiler techniques to overcome this challenge. And then in terms of data, satellite data are going to show me these palm oil plantations and mills both over space and where they're being developed over time. So to preview results a little bit, I'm gonna find that import tariffs are effective relative to the domestic palm oil tax. If it's the case that importers are coordinating and they can commit to long-term policy. Ah, but here, coordination and commitment are compliments, OK, and so we really need both of them. If we don't have coordination, we're going to have low coverage, which leads to low, uh, which leads to leakage and therefore to low tariffs. And then without commitment we can't uphold tariffs and so we're going to have low tariffs tomorrow. We therefore want to compensate with high tariffs today, but we can't under leakage because high tariffs are just pushing palm oil over to unregulated markets. So then what do we end up with, we end up with low tariffs tomorrow in addition to low tariffs today. So then to touch briefly on contributions, I have this new dynamic empirical framework for evaluating emission-based trade policy, and what I want to highlight here is that by focusing on one particular important industry, I'm going to have a pretty rich model of supply that's going to capture very important features like the dynamics and like the spatial heterogene. Now, second, leakage and commitment are independently well recognized and decently well studied, of course, in the very well studied, of course, in the environmental literature. Uh, I'm gonna combine them within a single framework and in doing so, be able to show how they interact. And that interaction is new, especially in an empirical setting. And of course speaking of empirics, I'm going to have these empirical estimates for palm oil, which is understudied, but and I don't want to understate this, a massive huge source of emissions that is important in and of itself if we want to hit our climate targets. So just to outline, I'll start with the setting. Then we'll go to the demand model which speaks to leakage, then the supply model, which is dynamic and speaks to commitment. And then for supply, let me highlight that the structural model captures that these investment decisions are forward looking. OK, and so future prices matter here. It's not going to be enough for us to just regress palm oil supply today on the price today. That's not all that farmers are thinking about. And then last, I'll run counterfactuals to quantify leakage and commitment. So maybe let me just spend 2 minutes going, uh, giving you highlights on the setting, and then I'll pause for questions. So, palm oil is a perennial crop. Once you grow the tree, it's going to keep giving you this fruit, OK? And this is what the fruit looks like. Uh, that fruit is going to be taken to a nearby mill, and it's going to become palm oil. It's going to get milled into palm oil. Of course, when we set up plantations to grow that palm oil, we're deforesting and destroying forest, and in this region, a lot of that forest is actually a special kind of forest called peatland forest. So here's a before and after, left and right. Peatland forests are forests that sit on swamps that contain deep layers of decomposing organic material called peat, and the trees would recover in about 100 years, but that peat would take 10,000 years to recover, very much non-renewable. And furthermore, that peat has 10 times more carbon than the actual trees themselves. Slash and burn is releasing all of that carbon into the air, and that's why palm oil emissions are so, so big and so significant. From there, palm oil is going to be used in many familiar products, both food like cookies and margarine, as well as non-food, detergents, soaps, and even cosmetics like lipstick. Now who's producing palm oil? It's Indonesia and Malaysia. They're producing a lot of palm oil and they're doing it for export. So they're 84% of world production and 90% of exports. For world consumption, Indonesia and Malaysia consume 20% domestically. And then the top three importers, the EU, China, and India, consume another 35%. So notice here that China and India don't really produce any, don't produce any palm oil themselves. So import tariffs here are really just a consumption tax with no discrimination needed based on where this palm oil is coming from. And just to wrap up this kind of settings section here I'm showing you how this industry is expanding over 3 decades over time. So the dark blue here of course is palm oil plantations, and what we see is as time unfolds, you know, as things unfold, we're seeing this massive expansion over time and over space, and all of that land sort of devastatingly is getting deforested. All of that peatland is going up into the air. In the form of emissions. So what am I doing with uh counterfactuals? I'm gonna roll back to 1988, put in regulation in the form of import tariffs, and then I'm going to roll forward and see how things unfold and change in response. So let me pause briefly for, for questions before we talk about the demand estimation. Great, if, if there are no questions, I'll just jump into the, this is, uh, this is Tristan, sorry, I was waiting for the moderator, but I'll go. Uh, so, uh, Article 2 of the Paris Agreement recognizes that countries have differentiated responsibilities and capabilities, uh, in addressing climate change, um, and you know this case of Indonesia is one where clearly, you know, you have a, a lower income country producing something that's carbon intensive. And this is a policy of rich countries, uh, to stop that. Um, I, I guess how, how would you account for an idea of differentiated responsibility, ah, in this model, uh, and you know, is there something you can say about that? Yes, thank you for the question. So differentiated ability is certainly Part and parcel of the approach here. This is kind of the motivation for thinking about these external external policies like importers as opposed to relying on the Indonesian government itself, which may have enforcement issues, uh, you know, at play. However, the differentiated responsibility, ultimately I'm, I will be able to speak to a bit. In particular, you're getting at this equity concern, right, where Indonesia and Malaysia happen to be endowed with certain natural resources that are potentially theirs to exploit, uh, particularly Indonesia as a lower income country. How can we be sensitive to that? At the very end, I'm going to show you what those equity concerns. Look like specifically because I have the full supply side specified, so I can quantify that producer surplus loss for Indonesia, for Malaysia in response to any kind of tariff that we put in place in counterfactuals. That kind of producer surplus loss is the kind of loss, the kind of damages that one would want to address with transfers in order to address this equity. Um, but you know, and one can also think about waiting, right? So when tariffs are set, they're set in my baseline analysis, waiting, you know, a dollar of producer surplus in Indonesia, 1 to 1 compared to consumer surplus in Europe, which is maybe. Unilevers $1 of consumer surplus, OK, uh, but if you want to upweight Indonesian surplus because maybe that's lifting people out of poverty, then in that case, maybe you want to upweight producer surplus, and, and that would change the kind of tariff that is, that is optimal. Hi, uh, this is Govindra Timilna. So it looks like you have a single commodity model, right? It's only for palm oil, but palm oil is used basically, you know, to produce the biodiesel. For example, even if there is an import tariff for palm oil, Indonesia, Malaysian farm, they can produce biodiesel, and they can export biodiesel, right? So in this case, the leakage still would be there. Yes, so, so the simple policy response is to also put the carbon tax on biodiesel, right? So not just palm oil, raw palm oil, but rather palm oil content. And, and so that's going to prevent that margin of leakage. Um, but the, the other thing I'll point out, and we'll see this in demand is, yes, this is one commodity, of course, but I'm not thinking about it purely in isolation. Right, because on the European side, for example, uh, palm oil is going to be a substitute with soybean oil, rapeseed oil, both of which the European Union produces a lot of, particularly rapeseed oil. And so demand, of course, is going to need to account for that, right? It's, it's not as if palm oil is the one and only good that people use and, and can't substitute to anything else. Go ahead. Please, please. Yes, so, uh, regarding the single commodity model, Indonesia clear field not just for palm oil. They clear field for all other crops as well. And how do you account for that? Are you going to say all the deforestation is all due to palm oil plantation, or there's some other way that you're going to measure specific for palm oil plantation? Thank you. So I, to be very upfront, I am not accounting for that in the kind of estimates that I'll show you today. And indeed this is going to be a source of carbon bias because, you know, you put in palm oil tariffs, palm oil production goes down, but if that land is still being slashed and burned for acacia, that's the number 2 crop in Indonesia, then there's going to be huge leakage on that margin, and then my carbon estimates will be, will be wrong. They'll be biased. Now the The reason I didn't put that into the model explicitly is because I can actually check that in the data. So I have data on acacia plantations, and I can look at substitution from palm oil to acacia, and that substitution turned out to be much smaller than I expected. I thought I was going to have a nested model where first we choose as a farmer between all of these different crops. Then if I choose palm, it's this exact model. If it's Acacia, then within that nest, it's again my exact model, and I thought it was going to have this cross nest substitution, but that substitution ended up being small, and the reason I think is that those other crops are all much less profitable than palm oil. The reason why basically all of the deforestation in Indonesia and Malaysia is happening for palm oil is because everyone is growing palm oil because it's the most profitable crop by far. And so part of this is just that, you know, Acacia is 7 times less profitable than palm oil. That means that not everyone who's no longer producing palm oil would move to Acacia, and so that's why that substitution, I measured to be small. Otherwise I could have had this more complicated model, but the substitution was small in the data, so I just didn't, didn't have that margin. Uh, I, I, I'm Malaysian. I must say that I'm very surprised that you associate all those, uh, development in Malaysia to palm oil plantation, which I know is going down, not going up in Malaysia. But this is what you, you show on the map, which is what much of my surprise. Yes, so in, in Malaysia, well, what's happened is it's been capped, um, and so the, the Malaysian government says you can't, you can't do more. Of course, Malaysia has, you know, it has more ability to enforce than than the Indonesian government. Um, but then also, you know, you can, you can put a cap, but then once you cut down the forest, it's all gone already anyway, right? So it's, it's, uh, there's no kind of going back on the peatland that, that has been destroyed. Um, but yes, I, I thought that there was going to be much more substitution. It, it turned out not to be, um, but that's, you know, That's in the data that that I observed. In some counterfactual worlds, maybe, you know, uh, you don't have the economies of scale in palm oil, and now Acacia takes off, uh, in that world, then I'm not going to quite be capturing all of that. Um, for that though, I would need a model of the Acacia industry, which isn't quite what I have here. But, but that's how we would, that's what you would need is this multi-industry. And again, I measured the substitution to be small. OK, so I'll just jump into demand. So for demand, I have an almost ideal demand system. I'm going to move through relatively quickly here just because I want to flag that the richness really is going to be on the supply side. But nonetheless, let's jump in. This is typical two-stage budgeting. First, consumers are choosing how much to spend on the vegetable oil category. That's that top line with the log log specification. And then second, consumers are allocating that spending between palm oil and other vegetable oils. I want to flag here that these substitutes, and this point came up a little bit before, these substitutes are not associated with big carbon emissions. So I'm not that worried about carbon bias, bias in my carbon estimates from this margin of leakage. Now the asterisk here is that South American soybean oil. does have some deforestation associated with it, with it, but I think that that bias is going to be small even there because one, it's only South American soybean oil. Second, Amazonian deforestation is really driven more directly by cattle than soybeans. And third, it's not peatland destruction. So we're also talking about an order of magnitude, order of magnitude less emissions than what we're talking about with palm oil. So then on the second line, omega is the palm oil expenditure share, Omega IT. It's a function of 1, a sort of secular oil specific time trend, 2, the prices of each oil product. OK, so I'm going to get on price elasticities as well as cross price elasticities, and then 3, the category budget. So what I want to flag here is that this is a product space demand system that's going to allow me to be very flexible on the cross-product substitution patterns. Now, capturing that substitution is precisely the advantage here relative to a reduced form approach, where, for example, if I just regressed palm oil demand on prices, even of course if I use instruments, then I wouldn't be fully accounting for that switching to these alternative products and therefore the demand elasticities that I would get would be, would be biased. So again, to think about leakage, I want those market specific demand elasticities. I'm going to get them by estimating demand by market. And then in terms of data, I have annual consumption by oil product and country. And then annual prices by oil. So of course you know prices are endogenous, price endogenity on the demand side. What do I want? I want a supply shifter. For that supply shifter, I'm going to use growing season rainfall shocks for foreign oil producers. For example, I'm going to instrument for US palm oil prices with rainfall shocks in Indonesia, which produces palm oil. So the bottom line here in terms of results is going to be that the importers, that is the dark blue lines, have relatively elastic demand. OK, and so what this means is that leakage is a problem. If China is unregulated, then because Chinese demand is relatively elastic, it means that it is going to expand and therefore it will offset EU tariffs. OK, leakage is an issue. On the other hand, actually, if importers coordinate, then leakage is not so bad. Leakage is not so bad because what do import tariffs kind of always miss, even if importers all coordinate? They're going to be missing domestic consumption in Indonesia and Malaysia, which isn't exported. However, this domestic demand turns out. To be pretty inelastic. That's the light blue line. And, and so leakage on that margin isn't that bad. And here I think it's, it's inelastic just because Indonesia and Malaysia produce so much palm oil that really palm oil is the main thing that is used. They're not really using too much olive oil, and so there's just less substitution available and therefore a lower demand elasticity. And then I can also aggregate up to give you the, I can also aggregate up to give you the world demand curve. So the 2015 curve is going to be farther to the right than the 1990 curve, that is to say, more demand at all prices, and I can trace out that intercept to give you the path of the rightward shift over time. And that's the graph on the right. So demand is rising as palm oil gets widely adopted, and note that this is in logs. So demand is rising fast over this period of time. This is why plantations have expanded so much and so quickly, as we saw on the map before. This is also why without regulation, because demand is going up and that's putting upward pressure on prices, this is why we would continue to keep expanding in the absence of regulation. Uh, so that's an answer. I, I take a quick pause for questions. Yes, yes, uh, David McKenzie, please go ahead and then Tom, uh, yes, so, so this was, uh, uh, just a quick clarification. When it's, when you're looking at the domestic consumption, is that picking up then, um, the use of palm oil as an intermediate input into these other products? And so the reason that it's inelastic may be because of the sort of export of some of these, these other products that are using palm oil. Yes, in this, in, in terms of the data, this USDA data, it's, it's measuring all of the consumption of the raw good, whether it be used ah directly ah for a final consumer product, as a final consumer product, or whether it be used as an input to other, other industries. Yes. Thanks for that. Please proceed. Great. Um, so, I will jump now into supply. So for supply, I have this dynamic model with sunk investment. So I'm going to divide land into sites, OK, and a site is a block of land that can invest in palm oil. And empty sites then are going to be potential entrants. OK, this is how I'm defining potential entrants in this market, and note that, you know, you can only have so many potential entrants because there's only so much land in Indonesia and Malaysia. Now, active sites are going to be operating, they're gonna be producing palm oil, they're gonna have one mill and some set of plantations that are up, up and running and online. So this is going to be a Hoppenheim style entry investment game with a dynamic competitive equilibrium. That is to say that sites are going to invest in deforest today in order to set up their palm oil trees, and then those trees are going to produce revenues in every future period. Now there's no exit here because, you know, once you've already paid the cost to set up the tree, marginal costs at that point are pretty low, so there's not really a big reason to exit anyway. Also, even if you exit, deforestation has sunk at that point, and that's really what I care about. And then where do tariffs enter? What's the point of tariffs? Well, tariffs reduce future revenues, and that in turn is going to reduce the incentive to invest and deforest in the first place today. Now, the structural model here, what is it buying for us? What is it capturing? It's capturing that future prices matter because these are forward looking investment decisions. So again, we don't just want to regress palm oil supply today on prices today because prices today are not the one and only thing that farmers are basing their investment decisions on. And furthermore, these are future prices, so it's really expected future prices that matter, not realized future prices, and the structural model is going to help me deal with those expectations in a flexible way and kind of specify the assumptions that I need to make on them. Now, in terms of the timeline, let me just show it to you graphically. This timeline is gonna summarize the choices. OK, again, we have the discrete and continuous elements here. So, so what are they? Well, we are an empty site, OK, that's, let's start on the left. Uh, we're gonna choose whether to build a mill or not. That's a discrete yes or no binary choice. If no, then we're in the lower branch. OK, and we're gonna have the same choice next period. If yes, then we're in an upper branch, in the upper branch, and then what we're going to do is we're going to pay that mill investment cost because we've chosen yes to build the mill. Then once you are there and you've built that mill, then you get to choose how much land to develop into plantations. On the intensive margin, and that's going to be a continuous choice. Should we develop 456, 6.5, you know, 6.25, whatever hectares of land into plantation. And of course we're going to pay that plantation investment cost. Big plantation, lots of land, bigger cost. That's now, you know, what's the point of having all of this investment costs that we're paying? Well, now starting in the next period, we're going to be producing palm oil and getting revenue. And then at a later point, we can also choose to expand our plantation. We don't need to develop all of the land at once. So a few comments here. First, these costs are going to be subject to logit cost shocks that we're going to realize before we make the decision, and we're going to base the decision on. Uh, Second, again, these are future revenues, so we need to be dealing with expectations. And then third, these two margins of spatial dependence, uh, these two margins are going to capture that spatial dependence between a mill and the land around it, OK, and, and also that big fixed cost for building the mill in the first place. So it's not like every hectare of land has its own cost function and its own cost shock. No, you know, there are groups of land that are working together as one unit, that's what these two margins are going to help capture. Now for the state variables, we have aggregate supply and aggregate demand which are observed and of course they're going to affect world prices, which in turn affect the revenues that firms are going to get and therefore affect the incentive to invest. So for supply, very simply, you know, many plantations online, high supply, low prices. And then for demand, demand curve far out to the right, high demand, high prices. So in terms of spatial he heterogeneity, which is very important here, sites are going to be spatially different in observed and unobserved ways. So there's observed heterogeneity here in palm oil yields, how much you can grow on any given plot of land, and also cost factors like distance to a port, which are important, a major port, which are important for transport costs. So that is to say that sites can be observably different. And why is this important? This is important because it's going to capture directly that this idea of the good land being developed first and therefore the supply elasticity of the land that is left is going to depend on what kind of land that is left, and probably it's the worst land if it's, it's the kind of land that's been left for a later period to develop. But there's a restriction here, OK, the restriction is that there is no unobserved site level heterogeneity above and beyond the logic cost shocks, which, you know, that's heterogeneity, of course, but it has a distributional assumption on it, so it's restricted. There's no site level unobserved heterogeneity, like a, a site fixed effect, if you want to think of it that way, because in this model, once a mill is built, it's built. So I don't observe this site level decision again and again over time. And so I can't identify that site level fixed that. Uh, Instead, what I need to do is I need to aggregate across space instead of over just over time. And therefore I'm restricting unobserved heterogeneity here to enter only at the regional level. Uh, so what that means is that I'm gonna have regional fixed effects and regional time trends. Uh, but let me note that that's still going to capture some very important unobservables, like, for example, political economy, where, you know, plantations in certain corrupt regions might be cheaper because the bribes are cheaper there, for example. I don't see that, but that's going to be absorbed by this regional unobserved heterogeny. And then in terms of data, I'm going to see these plantations and these mills, these plantation and mill investments over time and over space in the satellite data. I'm also going to see revenues. So I see world prices over time, and I'm going to have quantities from the yields data. So for yields, I'm using an agronomy model developed specifically for palm oil. This is a Hoffman et al. 2014, and then I'm going to be able to feed in climate data and get much more fine-grained than, for example, the FAO's agricultural suitability data, which, you know, many people use, but that's not quite what I'm using because it's not fine-grained enough for, for my purposes. Now, what I want to point out here in terms of the quantity decision is that here active plantations are just producing their yield in every period. So on the, the, the quantity decision, that is to say, isn't a matter of how much to produce on this given plot of land. Instead, that given plot of land is just producing its yield. OK, so the quantity decision instead is over how much land to develop in the first place, and again, I see that by satellite. And also cost factors that vary over space. So again, you know, the port distance thing, but also being close to major roads or major urban areas might be better because it lowers transport costs. Uh, and then I'm also gonna estimate how carbon stocks enter firms' cost functions. So what I'm gonna show you in a bit is that firms don't find emissions privately costly. That is to say that without regulation, they're just gonna keep on emitting because they don't care about those emissions. It doesn't enter their cost function. Uh it's a question by. Please go ahead, Mr. Alan. So the question is why, so I don't know if the government does it, but wouldn't it be natural for the Indonesian government to levy an export tax just to extract monopoly rents? Yes, so I, I am going to show you that counterfactual at the very end actually of the export tax. Ah, maybe we think that it's administratively easier to administer, you know, easier to administer because now you can just impose it at the port instead of, you know, 1000 different auditors across all of the forests of Indonesia. So I am going to show you that. Yes, Indonesia, because it has market power, will have this incentive to put in some kind of tax to, you know, a monopolist wants to reduce supply in order to drive up prices because they're a price maker. It turns out that that motive will impose some kind of tariff. That tariff is not going to be nearly big enough to match the size of the carbon externality that we're talking about today. So yes, that incentive is there, but it's small relative to the carbon externality. But then that tax is the entry point for the EU instead of trying to import. Why don't you give an incentive for the government to increase the tax? Yeah, so the very last counterfactual, I will show you this exact, um, this exact counterfactual, and then I'll show you why actually the, the Indonesian government, uh, isn't happy with that, that export tax as well. Once I have the graph up, it'll be a little easier to, to, to come to. So, so if I, I, I'll, uh, remember to bring this back up, but if I don't, please remind me. It's on the last slide. So, uh, Say also had a question. Hi, yes, um, so, couple of things. So, uh, first, so should we, you know, most of the production then of palm oil in Indonesia is, is, uh, you know, individual meals, or are there com you know, multinational, or are there companies basically that have multiple meals and so. So, so what, what share of the production is, is coming from, you know, individual, individual mills. And then, so, I mean, I guess related to that, I mean, should we think about these producers as price takers, uh, you know, given that most of the palm oil is produced in, in Indonesia and Malaysia, again, if, if there's, you know, a large firm, then, you know, the, you know, it's unclear that there would be. Uh, price takers, and then, you know, sorry, one last point. So, so should we, should we think about, um, so I guess, so there's, you know, in terms of production, there's no pests or other, uh, or, or, you know, is that, is that subsumed into, into the, into the climate shock, I guess, you know, if there's, you know, if there's, you know, uh, pests and stuff that might affect production, um, you know, as well, you know, yeah. Uh, so is it the case that climate is going to change production directly and are producers internalizing that? Um, that, no, because, uh, it's happening over such a long margin. Also, palm oil is pretty hardy a crop, so temperatures can go up a bit and it's, it, it will actually be fine. Um, but then are there pests, are there pests, and, you know, uh, that, that, that might, might, you know, yeah, might, might. You know, that damage the production of palm oil. There probably there are some. I'm sure I don't know the name of the specific pests that palm oil farmers deal with, to be upfront. So I don't, I'm not controlling for any of that directly, but that is going to be picked up when I estimate the cost structure of the industry, part of the costs of developing a given plot of land into plantation, operational plantation is going to be inclusive of pesticide, for example, to deal with these pests. Fair enough, fair enough. And then to get to your point on market power and market concentration. There are multinationals that have multiple mills. There are about 1200 mills in the data, but you know, um, not Sinar Mas, Wilmar International is a huge company, OK, one of the biggest palm oil traders. They actually only have 8 mills. OK, so they have 8, it's more than 1, but it's not 800. And so I'm assuming everyone is a price taker, and this is why we see that this market concentration here is pretty low. So the, the top. One producer in the world is FGV Burhad FGV Holdings Burhad. This is a Malaysian company. It's only 4% of world production. The top 20 producers are 27%. So it's a pretty unconcentrated market. This is going to be important because I have a substantive assumption that I need to make to have estimation be tractable, which rules out this, this market power. So that's, that's also why I'm just assuming everyone is a price taker. Um, in terms of, sorry, sorry, there are a couple more questions, Alan, uh, Aaron, then, uh, Ariana, then Harris. Try to both the questions and the answer is a little brief in the interest of time. Sure, so I'm going to like ask very briefly. So I think in the upper branch of the decision tree, you, you observe like planting more or like not planting every year, right? If that's the case, you can, you can estimate the unobserved heterogeneityity based on that like the Harold Sucher's bus engine replacement problem, right? So is that correct? That's correct. But then the problem is going to be that I don't see that until the mill is actually built. So for many places, I, you know, I have 30 years of data, but if the mill didn't get built until 2005, then actually I only have the tail end. So I do see my decisions, but it's not, and it's not balanced across all of the, all the mills. And so that's, that's why. And then at the, that's the intensive margin, the upper branch, but then for the mill itself, it's you can't do it at all. That's why I, yeah, that's why I just aggregated it to be regional for both. Thanks. I was wondering, yeah, I was wondering whether you can comment on on location specific production. Uh, so from what I understood, from what I understand, you said, the issue is the specific, uh, environmental conditions in Malaysia and Indonesia. And, and so I'm, I'm interested to understand whether the policy should be targeting a reduction in those settings and, and potentially an increase in other settings such as Africa, Gabon, and other places that are interested in, uh, in this area, or should we view it simply as an absolute reduction in production of. Yes, this is a very important point that you bring up. I, let me first say that I I am focusing on just Indonesia and Malaysia because that's where I have very rich data, and also they're 90% of production anyway, so I just call it a day there. Uh, but we can also think about kind of supply leakage to West Africa, South America. There is some production there. Nigeria, for example, is 3% of world production. But then even within Indonesia, there is this question of whether all palm oil is equally bad, and there the answer is no. So palm oil produced on peatland forest is 10 times worse than palm oil produced on non-peatland forest. Not all forest in Indonesia and Malaysia is peatland forest. And so then what we realized then is that this import tariff that I'm talking about, which treats in my baseline, I'm just going to treat all palm oil the same, good and bad palm oil. That's a blunt tool, right? Ideally, and that's not a Peruvian thing. Really, the Peruvian thing would be to differentiate between the good palm oil and the bad palm oil. So I do run that counterfactual. That is an important counterfactual. I don't set that as my baseline because when we're talking about import tariffs, actually it's much easier to just treat all palm oil the same, because otherwise, You need to know for this specific unit of palm oil, I need to trace back its production history over time and figure out where it came from, and I need to either do it myself or trust a certifier to do it correctly. Also, they're, you know, in the unregulated market, good and bad palm oil is a perfect substitute. So you have switching, you know, I'll take all of the good palm oil from China and send it to Europe because Europe says they want good palm oil. So it's not a kind of fix everything, but, but of course it will be better because it's targeted, and I do run that counterfactually. OK, final question for Harris, and then I think we should move on in the interest of time. OK, hi. So I have a question regarding how firms access the land. I mean, how much do they pay in order to obtain a lease? It's something that seems to be a little bit hidden in your model, maybe through the production costs. I mean there's no explicit modeling of the price of the land. I mean, there's been a lot of debate. About whether governments somehow squander the land by giving you very favorable access to our producers. So isn't there a possibility of having some policy through making land more expensive for producers rather than tax imports or exports. Yes, absolutely. So this would be a surprise. Well, I, I hear actually two things that you're saying. One is access to export markets. There, I'm just assuming that there's direct access and the world price is the world price. I'm abstracting away from that. But then at the level of, you know, land cost and production costs, that's all getting picked up. when I estimate the cost structure of the industry, one point that can be made is, you know, when I run my counterfactuals, I'm holding that cost structure fixed. OK, so if in response to palm oil tariffs, the Indonesian government is responding by changing the cost structure, I'm going to miss that. Although I can't get that without a model of how Indonesia kind of sets cost structure. But then also sort of, what about this policy of bidding up? What about Indonesia just bidding up the cost of land, for example, and reducing things in that way. There again, that's going to rely on an Indonesian government that is both willing and able to implement that policy. So one, does the Indonesian government actually want to do that given that palm oil is their number one agricultural export and would be their top export, full stop, if it weren't for the fact that Indonesia is also a big natural gas oil producer. Um, but 2, could the Indonesian government actually do it? So they could put in a fine for building in forested areas. The central government could do that, but then really because it's a decentralized government, actually it's going to come down to enforcement by the local mayor who, you know, I can bribe. Ah, and so there's an enforcement issue. This is why I'm focused on the demand side. These supply side policies are traditionally what environmental regulation has focused on, and I'm just pointing out that there's this one other qualitatively different kind of approach that might help in cases where all of that domestic regulation is very difficult and messy. Actually, we have another policy lever that we can use. OK, thanks for the clarification. Thank you. Great, so I'll just talk a bit now about estimation. So, for estimation, what I'm doing here is I'm combining classic continuous oiler methods with newer discrete oiler CCP methods, and both are going to be using short-term perturbations. So, I'm comparing investing today, which is what actually happens in the data, to delaying investment by one period, and by revealed preference, uh, that delayed investment is suboptimal because I didn't actually do it, and that delay is going to help me pin down the sub sub-optimality of that delaying is going to help me pin down the payoff parameters. So notice here in both cases, the investment is up and running by period T + 2. And so some under some assumptions, the continuation values are going to align and difference out. Ah, so differencing out the continuation values helps us a lot computationally because now I don't need to compute them. And that's the usual challenge. Also, future expectations are getting differenced out alongside the continuation values. That is to say, this is the sense in which I can accommodate those expectations without needing to specify exactly what they are, of course, subject to an assumption that I'm going to talk about in just one second. Um, so, what am I assuming? I'm assuming first that site owners are actually choosing between today or delay. That's gonna put some restrictions on, for example, property rights or credit constraints or uncertainty, although I can speak to each of those a little bit. Now I'm also assuming that sites are atomistic, and this gets at Xavier's comment from before, where this is going to rule out market power, OK, but this is a place where this is a setting where market concentration is low, maybe that's not that bad, but it's also gonna rule out spatial interaction, uh, which, you know, if I am a site that is neighboring you, another site, and now because I've developed my. Oil land into plantations. Now I'm attracting a local labor market, for example, and that's making it cheaper for you to produce. That's a very realistic, natural spatial interaction that I'm not going to be able to accommodate. Why? Because it's going to end up being hard to relax, one, absent a model of local labor markets and absent data on local labor markets, which I don't have. But two, It's going to be hard to relax because if I'm a big player and I delay, then others are going to respond to that delay, that I, my action, and that's going to change how the economy is evolving and where the state of the economy is going. That is to say, now the continuation values are no longer going to align and difference out, and then we're going to have computational issues. Now, lastly, I'm going to be assuming rational expectations, which by the way, does mean that expectations can be wrong today. They just need to be correct over time. Of course, still a very strong assumption, but it's going to be much weaker than needing to specify what exactly expectations are in every period, like we would do with a conventional full solution approach. Now, the upshot of this is we're going to be able to estimate the model with a linear regression, and so I can talk about endogeneity in instruments in familiar ways, and furthermore, Euler methods are going to accommodate the non-stationarity of the problem. So I actually can't use typical CCP methods like BBL or POB because those require stationarity. And if, you know, BBL and POB are just letters to you, then you can ignore that sentence, not a, not a, not central. OK. So let's start with uh estimation, uh, what, what I'm estimating, and let's start with the continuous choice on the intensive margin. Now, uh, if this is the plantation development decision, that continuous twist. This slide I want to flag is pretty standard. It's just like the consumption, consumption savings problem in, you know, grad school macro 10 1. So I'm gonna go through quickly, but I, I, I'm still gonna go through because the intuition is useful. So, in this top line, the Euler equation is capturing that trade-off between investing today or delaying. Now, this is a first-order condition. If I invest today, then revenues are higher because I'm going to start producing earlier, but at the same time, costs are also higher because I have to pay them today instead of later when they're discounted. So this is going to specialize to this second line. I'm going to develop more plantations today, AIT, uh, relative to tomorrow, AIT plus 1, for action, if the extra revenues are high or the extra costs are low. Now, note that that top line has expectations, but they don't show up in the second line because I'm substituting them out for realized values plus expectational errors. This is the usual trick. Uh, now, under rational expectations, those errors are mean zero conditional on anything in the period T information set. Note, of course, that we're talking here about uh RT + 1 and not RT, OK? And so we need to be a little bit careful. RIT + 1, not RIT. So we need to be a little bit careful here. I need to use the lags. So, let's then talk about this linear regression that we get, uh, and let's talk about identification. First off though, what are the data here? Uh, the left-hand side is data. OK, so I see in the satellite data how much plantation development A or action there is in each site I and in each period. AIT AIT plus 1, I see it. Beta discount factor, generically unidentified, you know, Manac Desmar, uh, you know. Not identified, I'm just gonna set it. I can play around with it though and show you robustness. OK, so the left-hand side, I see everything. And then prices P yields Y and cost factors like road distance, urban distance X. I see all of those, those are data. So then what am I actually estimating here? The goal is to estimate these parameters data of the payoff function. And then the error term is going to have two kinds of errors that we need to worry about. One, the unobserved cost shocks epsilon, as well as two, those expectational errors. Actually, both are going to cause us problems. So let's start with the epsilons. Those are going to cause price endogeneity for the following reason. Low costs today mean high entry today, so high supply tomorrow, and therefore low prices tomorrow. Now, the atas are also correlated for the reason I said before. This is PT plus 1, not PT. So for the instruments, I need to use lags. Um, OK, so price endogeneity on the supply side, what do we want? We want a demand shifter. So I have that demand shifter. I'm going to get that demand shifter from demand estimation, that's that right word shift over time in the demand curve that I plotted for you before. Of course, that's just time series variation at the end of the day. OK, so here what it's going to do is it's going to interact with that cross sectional variation in yields, which I can instrument for with exogenous factors like sunlight. So here's the intuition. When prices go up, it's going to increase revenues for sites that can actually grow palm oil. It's not going to help very much for a site that just can't grow palm oil to begin with. Why? Because, you know, revenue is price times quantity, so if quantity is zero, then revenue is always zero. Price going up doesn't matter. So that is to say that suitable sites are going to be more likely than unsuitable sites to develop land when prices rise, and that difference is going to give me the supply elasticity that I'm after. And then there's then there's a question about. I'm good. So yeah, I have two questions. Uh, first is, can, can you store palm oil at all? Uh, so, uh, palm oil can be stored. I don't have that either on the demand side or the supply side. On the demand side, side, I actually see the storage of palm oil. You might be worried about a stockpiling concern and dynamic demand. It turns out to be very small, and, and I, I see that it's small because it's in the data. On the supply side, wait one second. What does it mean it's small storage is, is, is non-zero, then, then you have. Then you have the hotening conditions kicking in. Yeah, but the thing is, if you look at the kind of ketchup papers or the laundry detergent papers, so Handel Nouveau, but also Michael Keene has work on this, there you can see they kind of back out, OK, for a given amount of consumption. Every day I'm consuming this much relative to consumption. How much do I have stored at any given period of time? And you see, I mean, we see this with olive oil that we just use at home, right? How much do we use today relative to how much is in the bottle? We're storing, you know, 3,000% of what we're consuming day to day. Here, because it's part, and part of this is just because of the aggregation, it's at the country level and it's year on year. You see that of the amount that is consumed every year, it's only like 10%, it's something like 10% being stockpiled at any given time. So this is the sense in which, you know, that stockpiling can go from 10% to 15%, but it's not going to buy us things in a way. I understand, but the fact that that the stockpile is non-zero and and the fact it's only 10% can be the outcome of. Of an economy which is driven by, by, by prices in the hoteling world. I, I, you know. The issue is, is as soon as there is a potential for speculative storage, then your price equation becomes pinned down by, by, by a different equation. Yeah, I agree. No, I agree. This is going to be a fundamental issue with at the end of the day, my demand is static demand, not dynamic demand. We can talk about making a dynamic demand. I'm not sure at the end of the day it's going to give me some price, some price elasticity, demand elasticity that's going to speak to the leakage question. We can talk about if dynamic demand would change that price. Elasticity from 0.7 to 7. If that's the case, OK, it's super important. Uh, but if it's going to go from 0.7 to 0.8, then it's not going to change qualitative conclusions by that much. But yes, I totally agree with this. No, I'm talking on the, I'm talking on the supply side. I'm talking not talking about demand in Europe. I'm talking about Indonesia traders or the stockpiling in order to, to sell strategically given the crisis that are going to happen. I think that's the most, most problematic problem because this is how you get your amplification. So your PT +1, YT plus one. Fully hedges on the fact that there is no, that, that your price opportunity is from a market clearing condition, not on the storage condition. Yes, yes, so I, I agree with that. That, that's, that on the supply side, I, I am missing that storage margin. Here again, so I, I, I don't have the, the data off the top of my head, but also this stockpiling within Indonesia is also observed. And so one track that I haven't done but should do, and, and thank you for Bringing it up is to see the extent to which that stockpiling is responding to the historical price variation, and I can start to at least speak to this more concretely in that way. Yeah, here, when you produce it, you have to sell it right away and, and therefore, yes, the, the price variation is just coming from, from, uh, you know, the, the, the kind of price variation period to period. Um, in counterfactuals, so that's an estimation concern. For counterfactuals, I'm not as worried, you know, conditional on the estimation being right, because here you might think that, you know, I try to tie, I see a tariff coming into play, and so I try to rush by selling all of my stockpile right now, for example, um, or I try to rush into putting, setting up a farm to kind of time out the tariffs being put in place. But here, tariffs are being put, are, are, are in place on consumption in every future period. OK, so there's kind of a limit to the extent. To which you can kind of rush and uh preempt the tariff in the way that you can in, in other environmental uh settings. No, no, I'm fine, but my only concern is, you know, when, when you showed your, your, your price trend, your log price that is very stable over time, it does look like a storage, uh, price pattern. Yes, I mean it's very constant returns. So, so that's why I, I, no, no, totally you make your reason not to have it, but that's what I wanted to try. That's it. Yeah, absolutely. No, I, I agree with that. Although the one kind of other point here is that if demand is just going up over time, as it is, you know, very quickly, then actually if you, if you expected this storage component to be really, really meaningful, and if palm oil can be stored for 567 years with no problem, as it can be. Then you would expect that everyone would just stockpile everything all of the time and never sell anything given that no, no, no, no, no, no, no, that's what the hotel and coalition tells you is people in every period are indifferent between selling and stockpiling. So it doesn't have any prediction on the quantity of stocks, but this pins down instead of having market con what I'm, my point is instead of having the market clearance condition to pin down your price in, in your early equation, it's a hotening condition that pins down the price. I see, I see that's a completely different, that's, that's a completely different early equation, yeah, because it's, it's, it's off of the storage versus selling as opposed to selling today versus selling tomorrow. Yeah, producing today and producing tomorrow. Yeah, I, I've coupled producing and selling. Your point is that producing and selling are actually two, when we decouple it, it becomes more complicated. You have this extra condition. Yeah, I, I agree with that. OK, great. So let's try to move on to the results, uh, in the interest of time because we have about 20 minutes left. Great. Um, OK, so let me talk to you about the extensive margin. So here, what's happening, uh, is that we have this, uh, mill being built. OK, so this is that discrete choice. Uh, so if this were the only margin, then we could just use the usual Euler CCP methods directly. So, let me Give you that intuition first. Either we can invest today or delay. Either way, the mill is being built by T + 2, so the continuation values align, we have finite dependence. Now, I am imposing that the mill actually gets built in that delayed case. That's potentially sub-optimal. It's going to depend on the loaded shock. So this correction term gamma is going to account for that. Now my model is a little different because we also have this continuous intensive margin. The problem here is going to be that delaying the mill also delays the plantation. If I delay and only plant AIT, then the continuation values don't align. And so instead, we need to catch up to that original path to get finite dependence. Again, I'm imposing a particular potentially suboptimal choice. The correction term needs to be expanded to account for that. So combining this continuous, these continuous and discrete pieces is the marginal methodological contribution here. This is common, a common setup. Businesses make a discrete choice to enter a market and then a continuous choice over how much to produce, but I don't want to overclaim here because CCP methods could already deal with both margins. It's just oiler CCP methods that focused on the extensive margin, and I'm relaxing that. So from here we're going to get a regression. This is a two-step estimator, just like the usual CCP methods. Without going into detail, what I'll just say is that this piece here is where the two margins hook together. OK, so this is where the intensive margin estimates come in. And what's the intuition here? It's just that I'm more likely to build the mill in the first place if those plantations would actually be good. If the plantations will be terrible, then why build the mill to begin with? OK, so now I'm going to have the cost structure of the industry which I'm going to need of course for counterfactuals. Here are those cost estimates, estimates. So I find that on average to develop one hectare of land into plantations, I get an average cost of $10,000. That's going to line up. With the $7000 accounting cost, different economic objects, of course, but we're in the same ballpark, so this is at least one way of starting to validate the estimates. Here cost factors don't enter because at this point the mill has sunk. You just kind of build the plantations around the mill. So for mills, I'm going to get a $23 million average cost compared to $20 million in accounting costs, and here the cost factors do matter. So being close to major roads, major ports, major urban centers, major urban centers is going to be less costly because you have lower transportation costs. Um, now we see here that sites actually do try to avoid forest. That's good. Maybe it's just because then they need to pay a bribe if they need to. Go and cut down all of these trees, but we see here that there's no accounting at all for peat. OK, so cutting, releasing peat emissions is completely inconsequential and not costly at all to these firms in terms of private cost, and that's where all of the big emissions are. OK. And so this is the sense in which private costs are failing to line up with social costs, and this is the externality that we want to force firms to internalize. OK, so let me put the pieces together for you now and we can talk about counterfactuals. So, suppose the EU puts in tariffs tar, conditional on supply, sellers are now going to be profiting less from the EU and so they're going to shift sales over to China. Now the demand elasticities are going to determine how much shifts over to China and therefore how much world prices fall. and lower world prices are going to translate into less supply given the supply. Now I have data on the carbon stocks, where the trees and the peat are, and so when supply contracts in this region versus that region, I'm going to be able to compute the implications for carbon reductions directly from the data. Now I am going to monetize emission reductions with the $40 social cost of carbon. And then, and this is getting at uh the question from before, so I'll not rehash everything. I am assuming that palm oil tariffs are not increasing non-palm deforestation. Acacia is the big threat there, but I check uh that substitution in the data, and it's, it's small anyway, so I didn't augment the model to account for that. Uh, otherwise, it would have been important to, to account for it directly. Now, the other thing that I'll just point out here is that, you know, sometimes this gets a little lost. This is an equilibrium model. You know, there's this leakage across markets. Prices are set in equilibrium, farmers are playing a dynamic competitive equilibrium, but this is only within palm oil. OK, so I don't have effects via other industries because that would require a multi-industry model. And so these trade computable general equilibrium models are going to do much better at capturing that cross-industry kind of leakage and those cross-industry effects, but the cost is going to be that the production functions within each industry and those models is going to be much more simple and a little bit crude compared to what I have, where I can really give you good detail within this one industry. OK, so in the baseline, I'm gonna set tariffs to maximize global welfare. I can relax that, and, and in that case, the terms of trade effects or the kind of Indonesian market power effects would come in. Uh, I do do that in, in robustness. Here I'm also to get at Ariana's question, I am treating all palm oil equally, again because it's much easier to administer, but I can and do relax that assumption as well. I can also differentiate between good palm oil and bad palm oil. So for leakage, I'm going to show that the effects of different, I'm going to show you the effects of different coalitions. So all importers, that's the big coalition or just the EU alone, that's the small coalition. I'm not going to be modeling the coalition formation game, but I am going to be able to quantify for you the incentives to defect. Uh, now for commitment, full commitment is tariffs upheld forever. There's no game here because the regulator moves once at the beginning and then producers follow. There's no back and forth. Uh, no commitment is just sequential static optimization. And then lastly, limited commitment is like a Chinese 5-year plan. OK, you set it for 5 years, you do it, and then you reset again, uh, after 5 years are over. And so, uh, a 2-year plan is very close to sequential static optimization because you're updating a lot. A 50-year plan is like full commitment because you don't update that much at all. So note here there is going to be this back and forth, there is a game because the regulator moves and then the producers respond and then the regulator moves again, taking producer's past response as given. So I'm going to need to solve by backward induction. Um, note here, and this sort of came up a bit before, but I'll just, ah, bring it up again to clarify, producers are atomistic and competitive, so there's no sense in which they're gonna rush to clear land and push tariffs to zero, because if you can cut down all of the land, then tariffs should go to zero optimally. But they can't do that because they're, they're atomistic. Uh, they also can't really rush to get in before the tariffs, and that's, and that's the green paradox because tariffs are getting paid in every future period. So you can rush and avoid tariffs for one period, but you're still paying it over the future horizon. Now, for the sake of time, I'm gonna skip how I, the technical details of how I actually solved the model. Let me just say this one point. So, when I estimated the model, I estimated it without actually solving it, because the Euler methods allowed me to do so. That meant two things. Excuse me. Number 1, computation was much easier, and number 2, there were fewer assumptions that I needed to make on expectations. Here, in counterfactuals, I actually need to solve the model. There's no getting around it. And so that, again, means two things. Number 1, computation is gonna be harder. I'm gonna need to use some computational tricks here. And 2, I am gonna need some more assumptions on expectations. I'm gonna need to actually specify them. Uh, but we can talk about that in if there's time for questions. So let me just get into the, the counterfactuals, the counterfactual results. So the baseline result here is that coordinated committed tariffs are effective, OK? How do we see that here? The Y axis is emission reductions, the higher is better, and the x axis is the commitment period. The solid blue line is coordinated tariffs. So committed coordinated tariffs are right over here, and we see they're pretty close to that, uh, domestic, socially optimal domestic tax, which I'm thinking of as infeasible, but will be our, our benchmark. So there are two caveats here. Number 1, there are going to be pretty big losses as coordination breaks down, down to the EU alone. So for the EU alone, for example, leakage concerns are going to kick in because the EU is acting alone and because it has low coverage, and that those leakage concerns are going to result in small tariffs. So losses here are disproportionate because low coverage is going to get compounded by low tariffs. Now, 0.2 is that there are big losses as commitment breaks down, and the starkest case here is no commitment, where tariffs never get applied because of time to build. So, the intuition here is that tariffs today are going to get set to zero because they can't produce, uh, they can't prevent new plantations, because there's no new plantations aren't producing anything that's taxable until tomorrow. So those new plantations are really only thinking about tariffs tomorrow, not tariffs today. But then by the time we get to tomorrow, those new plantations aren't going to, uh, by the time we get to tomorrow, tomorrow's tariffs are again going to be set to zero because from the perspective of tomorrow, the plantations that were new today are already sunk. OK, and so tariffs tomorrow can't change the fact that those plantations have been built, and so again, tariffs are going to be set to zero, that is to say they're always going to be zero. But there's a positive side here. So number 3, all of these policies are pretty efficient relative to other things like retrofitting power plants for carbon capture and storage. So unilateral EU action, yes, has a pretty small effect, 2 or 3% compared to, you know, 65%. percent in the, in that domestic intervention benchmark. So 2 to 3% is small, but first off, 2 to 3% is better than nothing. And then number 2, at that small scale, it's only $10 of surplus for gun per ton of carbon averted. OK, so pretty efficient. And then number 4, coordination is going to help us alleviate also the commitment problem. So let me rescale those blue lines for you. Under full coordination, that's that solid blue line, we see that committing for 5 years at a time is actually enough to achieve 95% of the full commitment outcome. And 5 years is salient because that's the political cycle. On the other hand, if the EU is acting alone, then it needs to commit for 20 years at a time to get to that 95% efficiency relative to the, the full commitment outcome. OK, and so this is the sense in which leakage and commitment interact. That is, as you make progress on coordination, you're alleviating the leakage problem and you're also alleviating this commitment problem. Now I can also zoom in on the division of surplus across markets, and here that exercise is going to show us that actually, unfortunately, coordination and commitment are both difficult in practice. So first off, there's going to be this incentive not to coordinate. So suppose the EU, China and India are all acting and all importers are and other importers here, Nigeria, Bangladesh, the US, etc. are thinking about joining the coalition. So if they do, then they're going to sacrifice, they're going to be sacrificing consumer surplus. We see consumer surplus goes down. They're happy to do so because it reduces emissions if they care about emissions. If they don't care about emissions and only care about consumer surplus, there's no point in sacrificing their consumer surplus and therefore no point in imposing import tariffs themselves. And furthermore, here they're going to be able to free ride because the tariff coalition in acting, that is the EU, China, and India in acting, are putting in tariffs and pushing down world prices. And so now these other importers compared to other worlds where no one is doing anything are actually even better off because they're enjoying lower, lower world prices. Uh, so this gap then is the transfer that we need to think about in order to induce participation, uh, into the coalition by these actors. So there's going to be a very similar incentive not to commit, where, you know, if you commit for a long period of time, that's increasing tariffs, and the point of doing that is more consumer surplus sacrifice for bigger carbon reductions. But again, that only makes sense if countries care about carbon reductions and carbon emissions. However, note that under coordination, actually this gap here is not really very big, um, but when coordination breaks down, so here the other importers have left, it's just the EU, China and India left, that gap we see is, is bigger. So again, this interaction between uh coordination and commitment. And then last, I can show you those effects on Indonesian and Malaysian surplus. So if the problems are only weak incentives for domestic regulation, then actually here tariffs can help a bit. So suppose importers threaten Indonesia and Malaysia. That is to say, either you put in the large socially optimal domestic tax or face tariffs. What we see here is that in most cases, you know, Indonesia and Malaysia are better off just accepting those tariffs, which are smaller and therefore hurt producer surplus by less. Uh, they'll just accept them. But if the threat on the other hand, is coordinated tariffs, then we see that actually Indonesia and Malaysia are better off just moving to that socially optimal domestic tax, which one isn't very much bigger than coordinated tariffs anyway. Coordinated tariffs are already pretty big. That is to say, in moving to this domestic tax, producer surplus is not going down by very much. However, now, Indonesia and Malaysia are able to get that government revenue from that tax that was before going abroad to Europe, you know, China, India, etc. Uh, now, we might also think, uh, and this came up before, that an export tax is a smaller ask that's maybe easier to administer. So administratively it's easier, as I said before, because now we only need 5 auditors at the ports instead of 1000 auditors to cover all the forests. Uh, but you know, again, the counterpoint is that maybe companies now only need to bribe 5 people instead of 1000 people, maybe it's easier. Um, but either way, uh, we, if we can put in the socially optimal export tax, actually we're going to achieve exactly the same emission reductions as these coordinated tariffs, because in both cases we're covering all imports. That is to say we're covering all exports out of Indonesia and Malaysia. Uh, but note, however, that these uncoordinated tariffs over here are also still not enough to push Indonesia and Malaysia to this export tax. Uh, and then also that export tax might increase, uh, as Orellana had brought up before, this palm oil production outside of Indonesia and Malaysia, in West Africa and South America, which import tariffs would kind of catch by default. Uh, so, lastly, note here, and this gets to Tristan's equity concern from the very beginning. OK, note that all regulation hurts Indonesia and Malaysia. Everything is negative across all of these policies. OK, and so, uh, if we have these equity concerns, which we should, of course, then at least one, I'm going to be able to quantify those damages for you, and therefore, 2, we can talk about the transfers that might be needed to address these equity concerns as, as much as we can. Uh, so, let me just take 30 seconds to conclude, and then we'll have some time for questions. Um, OK, so to summarize what we talked about today, we've talked about how import tariffs can be effective if they're coordinated and if they're committed. And this is going to help when domestic regulation is infeasible, including when those domestic issues maybe take time to change, but we want to act on the climate today and now. And then as we're setting, we look at the palm oil industry, which accounts for a massive 5% of global carbon emissions. But I want to end on a, on a positive note. Even though that past deforestation is sunk and it's over, there are still huge swathes of forests that remain intact, especially in Papua and also to a lesser extent on Borneo. And that's the scope of what we can still save if we go ahead and act and implement good policy today. Thank you. OK, that's great. Thanks for sticking to the time, Govinda. Sorry, I had another question. Alan, I'm just wondering about your CO2 reduction coefficient. Uh, the, if the palm oil is used for biodiesel, so in that case, the biodiesel substitute the fossil fuel and reduce the emission on the demand side. So whether you have accounted for that reduction in the emission in the supply side. Yes, so actually, I, I did that calculation. When you have biofuels, you're averting kind of fossil fuel emissions. It turns out that those fossil fuel emissions averted are very small compared to the kind of emissions needed to, to grow palm oil, uh, on average. Um, those palm oil emissions are so big that even if you account for that, it's, it's still, you know, it's on the order of 5%. Yes, you avert a little bit, but not nearly enough to account for the peat that you've destroyed. That's why Europe has removed, they used to subsidize palm oil for use in biofuels, and, and just in the past couple of years, they've removed those subsidies. Uh, and so it's just a little bit after my, my sample, but, um, they removed those subsidies because they realized, OK, actually, we really should not be subsidizing biofuel, uh, palm oil for use in biofuels. Not only that, the EU already banned use of palm oil for the biodiesel production starting in 2020. Yes, although it's, it's not a full-on ban, it's, it's phased out over time. Originally it was a full-on ban, and then, you know, Indonesia said we're going to cancel all of our Airbus, purchases, for example. And so there's diplomatic, there's a bit of a diplomatic row, and there was a WTO lawsuit, and so they shifted it to, I think it's by 2030 now. Everything's going to get phased out. Norway though, however, Norway's not an EU country, of course, but closely related. Uh, Norway just full-on banned and, and, and called it, called it a day. OK, that's great. If there are no further questions, then let's conclude the seminar and, uh, thank you very much, Alan. It was very interesting. Uh, thank you for meeting both.
showAllTranscripts
no
duration
PT1H7M26S
scene7File
worldbank/AllanHsiaoPalmOil
scene7Domain
https://worldbank.scene7.com/
scene7FileAvs
worldbank/AllanHsiaoPalmOil-AVS
title
AllanHsiaoPalmOil
description
AllanHsiaoPalmOil
showTimestampAndTranscript
yes
col-xs-12
col-sm-12
col-md-3
col-lg-3
col-xs-12
col-sm-12
col-md-7
col-lg-7
  • add-style
  • lp-body-content
Allan Hsiao (MIT) presented his research on the topic "Coordination and Commitment in International Climate Action: Evidence from Palm Oil" on February 2, 2021 as part of the Development Research Group Winter 2020-21 Seminar Series.
lp-heading-top-medium
lp-heading-bottom-medium
col-xs-12
col-sm-12
col-md-2
col-lg-2