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:208c2152-87c7-4344-9ac9-fc2544b3c394/play?assetname=Elite+Capture.mp4
keyFrameImage
timestamp

00:01 Um,

00:02 welcome,

00:03 everyone.

00:03 Thanks for coming on,

00:04 uh,

00:04 on a Friday afternoon to,

00:06 uh,

00:07 uh,

00:07 to hear a very interesting paper,

00:08 Elite Capture and Foreign Aid by our colleague Bob Rikers,

00:12 who is a,

00:13 uh,

00:13 senior economist in the Development Research Group in DEC in the World Bank,

00:17 co-authored with Jurgen Andersson and Niels Johansson.

00:20 Um,

00:21 you know,

00:21 this is a paper that's been in the news a lot lately,

00:24 and we all know that it's perilous to get information from headlines,

00:27 and so we thought that it would be very useful to have

00:30 a seminar in which,

00:31 uh,

00:32 Bob could actually present the entire paper

00:34 as opposed to sort of the,

00:36 the,

00:36 the headlines that have been making the rounds in,

00:38 uh,

00:39 in,

00:39 in Twitter and in the news and so on.

00:41 I,

00:42 I wanted to say,

00:42 uh,

00:43 just one other quick thing about the paper.

00:45 So,

00:45 The papers,

00:46 uh,

00:46 in the World Bank's working paper series and it has,

00:49 uh,

00:49 garnered 21,000,

00:51 20,

00:52 you're right,

00:53 21,000 downloads,

00:54 um,

00:54 since last Tuesday when it was posted.

00:57 This makes it the,

00:58 uh,

00:58 fourth most downloaded World Bank working paper,

01:01 um,

01:02 uh,

01:03 sort of in terms of lifetime downloads since 2014 when we started tracking it.

01:07 Now you might

01:08 wonder what the most downloaded World Bank working paper is.

01:11 And for those of you who doubt that the World

01:13 Bank has a well-developed sense of both humor and irony,

01:16 the most downloaded paper is a paper by

01:18 our colleagues Dorta Dumelund and James Trevino,

01:20 which is called Which World Bank Reports Are Most Widely Read.

01:25 Uh,

01:26 I,

01:26 I'm not making this up.

01:27 That is actually,

01:29 so,

01:30 anyhow,

01:30 so this all fits very nicely together and it's a,

01:32 it's a great pleasure to have,

01:33 um,

01:33 Bob here to present,

01:35 uh,

01:35 um,

01:36 uh,

01:36 to present about the paper.

01:37 Um,

01:38 because it's a large room,

01:39 we won't do our regular sort of seminar format of questions along the way,

01:43 but rather we'll hear first from Bob,

01:45 then we'll hear from our discussion,

01:46 uh,

01:47 Justin Sandifer.

01:48 Justin is the co-director of Education Policy and

01:50 senior fellow at Center for Global Development,

01:53 which is,

01:53 despite being a broad title,

01:55 extremely narrow compared to what Justin's.

01:57 Interests are and what his influence is in the developmental policy discussions.

02:01 He's worked extensively also on aid effectiveness,

02:04 which makes him a particularly apt,

02:06 uh,

02:06 discussant for this talk,

02:07 and he's really been,

02:08 uh,

02:09 a valued colleague in the Greater Washington area who I

02:11 always enjoy talking with and having him criticize my research.

02:14 So,

02:15 hopefully you'll do Bob the same courtesy.

02:17 And,

02:18 uh,

02:18 so Bob will talk for about 40 minutes,

02:20 maybe just in 10,

02:21 and then we'll open up for,

02:22 uh,

02:23 uh,

02:23 for Q&A.

02:24 So thank you again and over to you,

02:25 Bob.

02:34 Oh,

02:34 OK.

02:38 Good afternoon.

02:39 Thanks very,

02:40 very much for coming and,

02:41 uh,

02:41 for having me.

02:42 I'm here to present my paper,

02:43 Elite Capture

02:44 of Foreign Aid Evidence from Offshore Bank Accounts,

02:48 which is joint work with Niels Johannesen and Jurgen Juel Andersen.

02:51 But I would also like to use the opportunity

02:53 to thank uh our research assistant Nicholas Gomez-Para.

02:56 I'm not sure whether,

02:57 where he is

02:58 there,

02:59 like,

03:00 he's excellent and he's really made like

03:02 big contributions.

03:04 I also want to use the opportunity to thank many of you uh for providing comments.

03:08 If you feel that we haven't addressed your comment,

03:10 it's not because we're not taking it seriously,

03:12 it's probably because somebody else made a comment

03:14 suggesting to do the exact opposite.

03:17 So,

03:18 uh,

03:18 with that,

03:19 um,

03:19 I would like to start.

03:20 So,

03:20 a question that concerns

03:23 Certainly everybody in this room,

03:24 but I would argue the entire development community

03:27 is does aid reach its intended beneficiaries?

03:32 And the concern is that it may not,

03:33 that some part of it

03:35 may be captured by

03:36 those who need it least,

03:38 the elites.

03:40 And those concerns are certainly consistent

03:42 with economic theories of rent seeking.

03:45 And empirical evidence suggesting

03:49 That aid

03:49 doesn't necessarily flow to countries that are less corrupt.

03:53 And this

03:54 in some sense is pretty

03:56 intuitive

03:57 because if countries were amazingly well governed,

04:00 perhaps they wouldn't need so much aid

04:02 in the first place.

04:04 But of course,

04:06 this phenomenon is very difficult

04:08 to measure.

04:10 And this is

04:11 typical in the literature

04:12 on corruption.

04:14 One often has to rely on very indirect evidence,

04:16 and this paper is

04:18 no exception.

04:20 So the basic question.

04:22 We ask in this paper.

04:25 To address this concern

04:26 is

04:27 do World Bank aid disbursements

04:29 trigger increases

04:31 in foreign deposits in bank accounts,

04:33 he

04:34 in havens?

04:37 The

04:38 So what are havens?

04:40 Havens are

04:40 offshore banking centers

04:42 known for secrecy

04:44 and asset protection.

04:46 Such as

04:46 Switzerland,

04:48 Luxembourg,

04:49 the Cayman Islands.

04:52 And our main sample is going to comprise

04:54 22 countries.

04:56 That are really reliant

04:58 on foreign aid,

04:59 but World Bank aid

05:01 alone.

05:02 Accounts for more than 2% of GDP.

05:05 Each year

05:07 over our sample period,

05:08 which is from 1990

05:10 until 2010,

05:11 and our overall

05:13 aid.

05:14 is on average 10% of GDP.

05:19 But for context,

05:20 I want to emphasize from the beginning

05:23 that

05:23 these countries actually only absorb approximately 10% of all World Bank lending.

05:30 And an extensions

05:32 of the paper,

05:34 I'm going to show you sort of what has happened

05:36 in the period since 2009

05:39 when

05:40 there's been a significant crackdown

05:42 on,

05:43 on tax havens.

05:44 And what our results are for less aid dependent countries,

05:47 and I'm going to show you that in those countries we do not see this pattern.

05:51 Um,

05:52 that we see

05:53 in the most independent countries.

05:56 So the premise of the paper

05:58 is that haven deposits belong to elites.

06:02 How do we know this?

06:03 Well,

06:03 evidence from Scandinavia suggests

06:05 that

06:06 the top 0.01% richest households.

06:11 Own 50%

06:13 of the wealth

06:14 held in havens

06:15 and the richest 0.1%.

06:19 Own

06:19 80%

06:21 of that wealth.

06:22 And so there are similar studies in Colombia that sort of confirm.

06:26 That this offshore world is extremely concentrated

06:29 at the top of the

06:30 distribution.

06:32 By contrast,

06:34 The poor are typically

06:36 unbanked.

06:37 So thanks to the work of Ashley

06:39 and our colleagues,

06:40 uh,

06:41 in the finance group and DC.

06:44 We know that

06:45 in our sample of 22 aid dependent countries

06:47 in 2011,

06:49 Among the bottom 40% poorest households,

06:53 only 9%

06:55 had a domestic bank account,

06:56 so it's entirely implausible,

06:58 I would argue

06:59 that

07:00 the poor

07:01 are controlling

07:02 the money flowing to havens.

07:04 So that's the premise.

07:07 So what's our main finding?

07:08 Our main finding

07:10 is that

07:10 World Bank aid disbursements.

07:13 Do these highly aid dependent countries

07:16 coincide

07:18 with significant increases.

07:20 And haven deposits

07:21 within the same quarter.

07:26 And

07:27 How,

07:27 how big is this increase?

07:29 Well,

07:31 We define something we call the leakage rate,

07:33 which is the aid induced increase in haven deposit.

07:37 And that

07:38 Equals

07:39 roughly 75%

07:41 of aid inflows

07:42 for the average country.

07:45 But

07:46 because most of the aid actually flows to,

07:48 to countries

07:49 where

07:50 the stock of having deposits to GDP is relatively low.

07:53 As a share of all aid

07:55 going.

07:57 To these 22 countries,

07:58 this number is roughly 5%

08:01 and intriguingly,

08:04 This leakage rate tends to increase

08:07 with the ratio of aid to GDP

08:10 as I will show you later.

08:13 By contrast,

08:14 We do not find any evidence

08:16 that a disbursements trigger flows to non-havens.

08:20 Like

08:21 New York.

08:22 Or London.

08:24 Or

08:25 Gothenburg.

08:27 And

08:28 there's also no

08:30 evidence that this correlation

08:32 we document between

08:34 aid disbursements and the accumulation of haven deposits

08:37 has changed

08:38 over the last decade.

08:43 So I hope that by the end of the talk,

08:44 I'll,

08:45 I'll,

08:46 I've managed to convince you that sort of this,

08:47 this correlation exists and is very real.

08:50 What I think is,

08:51 is much harder to figure out

08:53 is what is driving

08:54 this correlation that we document.

08:58 And that's because we simply don't know

08:59 who owns these deposits because we have macrodata

09:02 and because we cannot track individual financial flows,

09:05 it's very difficult to attribute

09:07 agency.

09:10 Nonetheless,

09:11 I'm going to argue that elite capture

09:13 is an explanation that's certainly consistent

09:17 with the totality of the patterns

09:19 we observe,

09:20 but

09:21 Other explanations are possible

09:24 with our data,

09:24 unfortunately,

09:25 you're not going to be able to prove or disprove

09:27 any of them.

09:28 And if you've any ideas or suggestions,

09:30 I would very,

09:31 very,

09:31 very much welcome them.

09:34 Uh,

09:35 but before proceeding,

09:35 I think it's important to highlight

09:38 some of the important caveats.

09:40 Uh,

09:41 of our analysis

09:42 and

09:43 you know some misperceptions that,

09:45 that I think have made headlines.

09:47 So

09:48 for instance,

09:49 for instance,

09:49 according to the Times,

09:51 Billions of foreign aid disappears into tax havens.

09:54 Yes,

09:55 quite possibly based on our analysis,

09:58 but correlation doesn't necessarily imply causation,

10:01 so we cannot be confident

10:03 that

10:04 this is what is driving our results.

10:06 And on top of that,

10:07 money is fungible,

10:08 so it's not clear that

10:10 if diversion is taking place,

10:12 that diversion is out of aid flows as opposed to,

10:16 say,

10:17 money that was earmarked

10:18 for the Ministry of Education.

10:21 Second,

10:22 some newspapers have,

10:23 have sort of

10:24 made

10:25 what I

10:26 deem relatively grandiose

10:28 claims about sort of the,

10:30 the magnitude of leakage we document.

10:32 So for instance,

10:33 according to Forbes,

10:35 as much

10:36 as like 15%

10:39 of foreign aid

10:40 flows into uh tax havens.

10:43 Uh,

10:43 I think this is a wild exaggeration

10:45 because again,

10:47 Our main south pole

10:48 comprises 22 countries that together absorb

10:51 10%

10:52 of all World Bank lending.

10:54 We find a leakage rate of roughly 5%.

10:56 We don't find.

10:58 Leakage sort of in less aid dependent countries,

11:01 so expressed as a share of the World Bank's overall portfolio,

11:05 the true number is

11:06 much likely to be closer to 0.5%.

11:09 So,

11:09 so it's not the case that sort of

11:11 this pattern

11:13 applies across the board,

11:14 which I think is very important

11:16 to bear in mind when we discuss

11:18 our results.

11:20 So this concludes the introduction of the paper.

11:22 Next I'll tell you a little bit about our data,

11:24 our empirical strategy,

11:26 uh,

11:26 the various robustness tests we've done,

11:29 and I would like to spend more time on the interpretation and

11:32 this is also something where,

11:33 you know,

11:34 I would welcome your,

11:35 your,

11:36 uh,

11:37 feedback because many of you have,

11:38 have expertise that I think is extremely relevant to interpreting

11:42 these findings

11:43 and then I'll wrap up.

11:45 So

11:45 our main data set is a data set of cross-border

11:48 bank deposits,

11:50 sorry,

11:51 non-bank deposits,

11:52 so it excludes

11:53 financial

11:55 institutions,

11:56 banks,

11:57 including the central bank.

11:59 Uh,

12:00 so you can think of these as privately held account

12:03 firms and individuals,

12:05 but we only observe

12:06 immediate ownership

12:08 of these.

12:09 This data is confidential and comes from the

12:12 locational banking statistics

12:14 of the Bank for International Settlements,

12:17 which covers,

12:18 uh,

12:19 bank deposits and roughly 43 financial centers

12:21 owned by residents of around 200 countries.

12:25 And so we use confidential data that covers

12:28 all these uh

12:29 financial centers

12:31 for the period 1919 to 2010.

12:33 But then we also replicate our analysis with the publicly available BAS data

12:38 which allows us to also examine what happened

12:41 over the last

12:42 decade.

12:43 But these data have much more partial

12:45 coverage because they're based on sort of voluntary reporting of,

12:48 of countries.

12:50 Um,

12:51 and our key dependent variable is going to be

12:54 the sum of all haven deposits owned by residents

12:57 of a particular country

12:59 in the 17 havens in a particular quarter.

13:01 So think about sort of

13:03 All the money

13:05 that

13:06 residents of Kyrgyzstan

13:09 have

13:09 in Switzerland,

13:11 that they have in

13:12 Luxembourg,

13:12 that they have in Belgium,

13:13 like any of these 17 havens,

13:15 and they were going to add

13:17 these up.

13:19 For foreign aid,

13:20 we use information

13:22 on

13:23 project level disbursements

13:24 from the World Bank project database.

13:27 Now there's a real irony in this work because the reason we focus on the World Bank

13:31 is that it is in fact the only

13:34 major donor

13:35 for which we can construct

13:37 a series at the quarterly level

13:40 for a long period of time.

13:41 So

13:43 because of the transparency,

13:45 In our institution,

13:46 I think we're now under scrutiny.

13:48 Um,

13:50 so,

13:50 so in this data,

13:51 we know for each project,

13:53 project when it was approved,

13:54 how much money was committed,

13:55 what sector,

13:57 uh,

13:59 The project was in,

14:00 the type of lending instrument

14:02 and the timing of each disbursement,

14:04 and this is going to be extremely useful

14:06 when we try to

14:07 Address potential endogeneity of it because we can use this

14:11 to predict.

14:13 How much aid

14:14 we

14:14 would expect to flow to these countries

14:17 if

14:18 Aid would flow sort of typical

14:20 patterns.

14:21 So,

14:23 Our measure of aid is simply the sum of

14:24 all these project level disbursements on iodine IBRD projects

14:28 in a given country in a given quarter.

14:32 So to be clear,

14:32 this excludes

14:34 debt relief and it also excludes

14:36 uh

14:37 trust funds.

14:40 In addition,

14:40 we,

14:41 we,

14:41 uh,

14:41 use a number of,

14:42 uh,

14:42 other data sets,

14:43 so

14:44 we're going to exclude periods of war,

14:46 coups,

14:47 crisis,

14:47 disasters,

14:48 uh,

14:49 we look at how the results vary with petroleum,

14:52 rents,

14:52 financial sector development,

14:54 the capital account,

14:56 and disclosure requirements for politicians.

14:59 Uh,

15:00 you're also gonna look how it varies with

15:01 democracy,

15:02 uh,

15:03 corruption,

15:04 and

15:05 importantly with sort of the share

15:08 of aid that is procured domestically

15:10 versus

15:11 internationally.

15:12 Because

15:12 you might think that sort of these patterns that we're documenting

15:16 are just an artifact

15:17 of,

15:18 of contractors who execute a sponsored contracts.

15:24 Depositing some of their profits

15:26 in in havens,

15:27 and

15:28 I'm going to argue that,

15:29 yeah,

15:29 maybe that's,

15:30 maybe that's happening,

15:32 but it's quite unlikely.

15:35 So,

15:37 Just to give you a sense that these are distributions of aid

15:40 as a share of GDP

15:41 and,

15:42 and

15:42 even deposits growth.

15:44 As you can see,

15:44 there's quite a lot of variation in in aid inflows,

15:47 which is very useful for identification.

15:49 And you also see,

15:51 like,

15:51 when you look at the distribution of haven deposits growth,

15:54 that in many quarters,

15:55 they actually go down.

15:56 So,

15:56 so there is this saving,

15:58 which I think is relevant to bear in mind.

16:01 So,

16:02 and some descriptive statistics on,

16:04 on,

16:04 on the sample.

16:06 So World Bank aid in the sample accounts for roughly 2.7% of GDP on average.

16:11 Overall aid,

16:12 uh,

16:14 It's roughly 10% of GDP.

16:16 So these are economies in which

16:17 aid is really

16:18 a big deal.

16:21 Uh,

16:22 when we look at the level of haven deposits,

16:24 we see that

16:25 It stands on average around 61 million

16:29 and the level of haven deposits stands around 136 million.

16:33 So this I think is an extremely

16:35 relevant descriptive statistic

16:38 because it suggests that

16:39 On average,

16:42 It would seem that residents of these countries

16:44 have a strong preference for depositing money in non-hans

16:48 relative to havens,

16:49 so that makes it all the more puzzling

16:51 that when aid flows in,

16:52 we only see outflows.

16:54 To havens and not to non-hans,

16:56 so it's not simply not true

16:58 that,

16:59 you know,

16:59 residents of these countries routinely

17:02 Prefer depositing money

17:04 in havens.

17:07 Uh

17:11 So,

17:12 our empirical strategy is extremely

17:14 simple.

17:15 We're going to

17:16 regress the growth in haven deposits measured in logs

17:20 on aid

17:21 as a share of GDP.

17:23 Controlling

17:25 for

17:26 GDP growth

17:27 and in robustness checks,

17:28 we also had

17:29 controls for

17:32 Um,

17:32 the mechanical

17:34 appreciation

17:35 in

17:36 deposits that would result from,

17:38 from exchange rate

17:39 movements alone.

17:41 As well as commodity price shocks because you might

17:43 expect that sort of some of the wealth that's going into havens

17:47 is the result of

17:49 these commodity price,

17:50 price movements.

17:52 Uh

17:54 We're going to control also for country and time fixed effects.

17:57 So effectively we're asking whether

18:00 a country

18:01 when it receives aid.

18:03 Accumulates deposits more rapidly than in times when it doesn't.

18:09 Uh,

18:09 standard errors are going to be clustered at the country level,

18:11 which is conventional,

18:13 and we're going to Windsorize our deposit and aid measures,

18:16 uh,

18:17 at the 1% level

18:18 because there are some extreme outliers,

18:20 but these are this Windsorization

18:22 is not

18:24 driving the results as I'm gonna.

18:26 show you.

18:27 And then in alternative specifications,

18:28 we also look at the growth in non-haven deposits

18:31 and the differential growth of haven

18:33 deposits relative to non-haven deposits.

18:35 So if you thought that,

18:36 you know,

18:36 this was just a story about sort of

18:39 Aid sort of stimulating

18:41 economic performance and

18:43 leading to capital

18:44 outflows,

18:46 that wouldn't necessarily explain why there's differential,

18:48 uh,

18:49 growth

18:50 outflows to havens versus non-havens.

18:52 But of course,

18:52 a major problem here is aid is deeply endogenous.

18:57 So it's quite likely that,

18:59 you know,

18:59 a country receives aid at a time

19:01 when it needs it most,

19:02 for instance,

19:03 when it's experiencing

19:05 A natural disaster

19:06 or a financial crisis,

19:08 when I would argue it's also particularly appealing to

19:10 take your money out of the country and park it

19:13 somewhere else.

19:15 Um

19:16 So

19:19 We're going to address this in 180 in three ways.

19:21 First,

19:22 we're just going to augment our specification with both

19:25 lags

19:26 and leads of our aid variable

19:28 to make sure that

19:29 it's really

19:30 aid

19:31 that is driving

19:32 the surge in haven deposits

19:34 so that there are no pre-trends.

19:36 Secondly,

19:36 we're going to use an instrumental variable

19:38 strategy,

19:39 drawing on an excellent paper by Art who had a simple but powerful insight that

19:44 how much aid a country receives in any given quarter

19:47 is to a very large extent determined by past commitments

19:51 that cannot respond

19:53 to sort of contemporaneous events,

19:54 contemporaneous

19:56 shocks.

19:56 And so if we take those past commitments

19:59 and sort of we apply sort of the typical disbursement

20:02 schedule that applies to World Bank loans.

20:04 Then

20:05 we can come up with sort of a synthetic measure of predicted a disbursements

20:09 that's plausibly exogenous that we can use as an instrument.

20:13 And the third,

20:14 we're going to exclude

20:15 the episodes of turmoil.

20:18 Um,

20:19 such as war,

20:19 coups,

20:20 uh,

20:20 conflict,

20:21 and financial crisis.

20:23 So,

20:24 these graphs show you the first stage of our

20:27 instrument with aid

20:29 residualized on the left hand side and predicted aid on the right hand side.

20:33 And as you can see,

20:33 the correlation is,

20:34 is really strong.

20:36 So,

20:37 as I already sort of explained,

20:38 it's based on sort of predicting project level quarterly disbursements

20:43 based on

20:45 sort of the typical

20:46 disbursement profile

20:48 for

20:49 Projects in the same sector,

20:51 in the same region,

20:52 executed in the same time period.

20:55 And then simply aggregated those predicted disbursements

20:58 to get a measure of sort of

21:00 country level

21:01 predicted World Bank

21:03 aid.

21:05 Um

21:07 So on to the main result,

21:09 so.

21:12 This table shows you our,

21:13 our main results,

21:14 so.

21:15 It shows that

21:17 When

21:19 It

21:20 increases.

21:22 The growth in haven deposits accelerates.

21:25 In fact,

21:26 if aid goes up by 1%,

21:27 the growth of in haven deposits

21:29 accelerates by roughly 3.4%,

21:34 and we do not see a similar

21:36 response

21:38 of non-haven

21:39 deposits.

21:41 So

21:42 the differential which is presented in the 3rd column.

21:45 Between haven versus non-haven deposits growth is also statistically

21:49 significant.

21:51 I would also like to use the opportunity to point out

21:53 that when we look at the coefficients on GDP growth,

21:56 those are much more similar

21:57 for both haven and non-haven deposits growth.

21:59 So

22:00 they're 0.12 and

22:01 0.15.

22:03 So it seems that both haven deposits and non-haven deposits responded relatively.

22:08 Symmetric proportional

22:10 fashion

22:11 to,

22:11 to sort of general increases and,

22:13 and or decreases in economic

22:16 um

22:16 activity.

22:18 So sort of this,

22:19 this asymmetric response of haven deposits

22:22 seems to be very particular

22:24 to,

22:24 to aid.

22:28 But you may be concerned that perhaps sort

22:29 of there were earlier events that triggered both

22:32 uh

22:33 aid inflows

22:34 and,

22:34 and the accumulation of having deposits.

22:37 So what this graph here does,

22:38 it

22:39 sort of shows you.

22:41 Um,

22:41 the results of

22:43 a specification in which we add lags and lead.

22:46 And so

22:47 the quarter relative,

22:49 the,

22:49 the

22:51 Dots are sort of point estimates

22:53 for

22:55 the coefficient estimates of the lags and leads

22:58 and the,

22:58 the bars are the confidence intervals.

23:01 And as we can clearly see,

23:02 sort of we see a big spike.

23:05 On

23:06 At the time the aid arrives,

23:08 but no evidence sort of

23:09 of responses to

23:12 aid received in earlier quarters or to

23:15 aid that's about to,

23:16 to come in.

23:18 Uh,

23:19 this is the same graph for non-haven deposits which don't seem to significantly

23:23 respond

23:24 to aid inflows,

23:25 although they're in a small,

23:26 insignificant dip,

23:28 uh.

23:29 At times 0

23:31 and of course then the difference.

23:34 Between

23:35 the accumulation of haven deposits and non-haven deposits

23:38 is also

23:39 Yeah.

23:41 Exhibiting a very similar pattern

23:43 as the one we saw two slides ago.

23:47 Yeah

23:50 And

23:50 how robust is this result?

23:52 Well,

23:52 when we instrument aid,

23:54 What we see,

23:55 which we do here in columns 2 and 3,

23:57 where the difference between

23:59 the 2nd and the 3rd column is that we

24:01 exclude sort of more of the initial

24:03 quarters

24:04 to,

24:05 to have sort of uh ar more exogeneity.

24:08 And you see that sort of the coefficient drops

24:10 somewhat,

24:12 as you would expect,

24:13 but it remains.

24:16 Similar in magnitude,

24:17 I would argue,

24:18 and

24:19 Strongly statistically.

24:22 Significant.

24:22 So this should sort of

24:24 assuage a lot of the concerns about

24:26 endogeneity.

24:28 Nonetheless,

24:28 we also examine what happens when we exclude periods of wars,

24:32 coups,

24:33 disasters,

24:33 and financial crisis,

24:35 uh,

24:35 and you can see that

24:36 our main finding.

24:38 holes in the coefficient

24:41 remains similar in size and also strongly statistically

24:44 significant.

24:44 And it's also true when we sort of exclude all of these

24:47 at the same time,

24:48 which I'm not presenting here.

24:51 Then we also run a number of alternative specifications.

24:53 So

24:54 crucially,

24:55 in column 5,

24:56 we

24:57 include country year fixed effects.

24:59 So this is an

25:00 extremely demanding test,

25:01 I would argue,

25:02 because now all the variation.

25:04 is coming from

25:06 looking at what happens

25:07 to that same country.

25:10 Uh

25:13 In the same year,

25:14 Uh

25:16 And

25:16 as expected,

25:17 sort of the results are slightly less significant.

25:19 They remain significant to the 10% level,

25:21 but the coefficient estimate is very,

25:22 very similar.

25:24 In column 6,

25:25 we

25:26 include exchange rate control.

25:28 So some people have said,

25:29 oh,

25:29 perhaps

25:30 this is just something to do with sort of how exchange rates move.

25:33 Well,

25:33 here we

25:35 sort of control for the mechanical appreciation

25:38 of deposits that would

25:40 eventuate because of exchange rate movements

25:43 alone.

25:44 This

25:45 does not seem to meaningfully alter

25:48 our estimated coefficient.

25:50 Then in column 7 we control for resource dependence and

25:54 fluctuations

25:56 in the price of oil.

25:58 Uh

25:59 By interacting

26:01 A dummy for being a resource,

26:03 uh,

26:03 being an oil producer

26:05 with,

26:06 uh,

26:06 our time dummies,

26:07 and the reason for doing this is that my co-authors have a similar paper

26:12 where they examine what happens to

26:15 resource windfalls

26:17 and they show that

26:18 in autocratic countries roughly 15%

26:21 of

26:22 um

26:23 oil wealth

26:24 seems to be flowing

26:25 into

26:26 tax havens

26:27 but

26:28 controlling sort of for this doesn't really

26:31 Impact,

26:32 uh,

26:32 our estimates of the impact of aid.

26:35 And then finally,

26:36 in the final column,

26:37 we show what happens

26:38 when we just use the raw data and don't wins our eyes and

26:42 as you can see,

26:42 the results are strikingly similar

26:45 to uh our main results.

26:49 So I hope that by now you're

26:51 somewhat convinced that sort of this pattern is real,

26:54 but to understand the mechanism,

26:55 I think it's important sort of to look

26:56 at

26:57 how this

26:58 pattern

26:59 varies

27:01 across countries.

27:02 Now,

27:03 it's important to bear in mind that our sample is extremely small.

27:06 We only have 22 countries,

27:07 so

27:08 naturally we don't expect sort of

27:10 Huge differences,

27:12 but nonetheless sort of to see whether

27:14 at least sort of

27:15 qualitatively these,

27:16 these parents could be suggestive.

27:18 What we do is sort of we split the sample.

27:21 Every time

27:22 into countries that sort of

27:24 have above and below the median.

27:27 Control of corruption,

27:28 for instance.

27:29 So here we see that.

27:32 Outflow is slightly larger,

27:34 but certainly not significantly larger

27:36 in countries where the

27:38 control of corruption

27:40 is lower.

27:41 We also see that outflows are slightly larger,

27:44 but again not significantly larger

27:46 in countries that require their politicians to disclose their assets,

27:50 i.e.,

27:51 countries

27:51 where

27:53 You could argue politicians should have an incentive to,

27:56 to use

27:57 havens.

27:59 The fact is also slightly

28:01 higher in countries that are less democratic.

28:04 Uh,

28:05 and

28:07 In countries

28:10 With better developed financial systems.

28:13 And last but not least,

28:16 The effect

28:17 is slightly higher

28:18 in countries.

28:20 That

28:21 Procure

28:23 more

28:25 of

28:27 They

28:28 um.

28:30 A larger share of the 8 contracts

28:32 from foreign firms.

28:33 So if you had expected that sort of this was a story

28:36 about sort of domestic contractors depositing.

28:39 They're legitimately earn profits

28:41 in havens.

28:43 You might not have expected.

28:45 The,

28:46 the opposite and that's not what we see.

28:48 Now you could argue that,

28:49 well,

28:49 you know,

28:50 these contractors must subcontract,

28:51 so

28:52 you know,

28:53 whether you give it to a foreign firm or

28:55 uh a domestic firm in the end,

28:57 like they're all going to rely on the same domestic subcontracts

28:59 that I think is like a legitimate contract critique sort of to,

29:02 to this

29:03 point.

29:03 So,

29:03 so I don't think like,

29:04 you know,

29:05 anything I'm showing is sort of

29:06 uh conclusive,

29:07 but I do think it,

29:08 it raises questions.

29:11 And third,

29:12 I want to look at some of the type of aid,

29:14 so.

29:15 Uh

29:17 Based on a lot of conversations sort of with country economists,

29:19 my prior had been that

29:21 if we believe that these patterns are the product of diversion,

29:23 we should perhaps expect to see

29:25 larger effects for development policy lending,

29:28 where there's relatively less

29:30 oversight,

29:31 and that's actually not what we see

29:33 in the data.

29:34 We see,

29:34 if anything,

29:34 slightly stronger effects for projects.

29:37 But then,

29:38 uh,

29:38 a colleague in the integrity vice president,

29:40 he suggested that this is exactly precisely what he would have expected

29:44 because his argument was that,

29:45 well,

29:45 most of the money is actually going through projects

29:48 and if you want to capture these funds,

29:49 you typically do it through politically connected firms.

29:52 Now,

29:52 whether that's a,

29:53 uh,

29:55 plausible argument,

29:55 I think it's a matter of,

29:56 of debate,

29:57 but

29:58 I frankly do not have a

30:00 completely compelling explanation as to why

30:03 we see these differences,

30:03 but I should emphasize that these differences are extremely small and not

30:07 Statistically

30:08 significant,

30:08 so I think it's also important not to overinterpret them.

30:13 Uh.

30:16 So

30:18 then we replicate our analysis using publicly available data

30:22 to see whether the crackdown on tax havens that we've been seeing in the past

30:25 10 years since 2009.

30:28 has sort of had an impact

30:30 on the patterns we see

30:31 in the data.

30:33 And

30:34 basically the coefficient estimates we get for

30:37 The period

30:38 from 2010 to 2019 relative sort to a free period.

30:43 Coefficient estimate is very strikingly similar in magnitude.

30:47 But as you can see like the last period like has a much wider confidence interval

30:52 so

30:53 you know I don't think we can

30:54 conclusively conclude sort of that transparency hasn't,

30:57 uh,

30:58 helped.

30:59 But

31:01 The point estimates

31:03 suggest that the change

31:05 has been very limited.

31:09 OK,

31:09 so

31:10 how big is sort of the aid induced

31:13 increase in offshore deposits,

31:14 what we refer to

31:15 as leakage?

31:16 Well,

31:17 this is not something our model uh

31:19 delivers directly.

31:20 So we have to rely on a simple

31:22 transformation.

31:23 So the coefficient we get on aid is 3.4.

31:26 So it means that if aid goes up by 1%,

31:29 growth in haven deposits accelerates by 3.4%.

31:35 On average,

31:36 The stock of haven deposits

31:39 express as a share of GDP in our sample

31:41 is 0.22.

31:44 So

31:45 when we multiply 3.4

31:47 with 0.22,

31:49 we get 7.48 or

31:50 let's say roughly 7.5%

31:53 uh.

31:54 Is the implied leakage leakage rates because if

31:56 aid goes up by $1

31:58 haven deposits

31:59 should go up

32:00 at the sample mean by

32:02 $7.5.

32:04 But

32:05 this

32:07 Calculation applies to

32:09 the average country,

32:10 but it happens to be the case that most of the aid is actually flowing to countries

32:14 with a relatively low

32:16 stock of haven deposits to GDP.

32:18 So if we do the same calculation using the same 3.4%

32:22 number applying,

32:23 assuming it applies uniformly.

32:26 But

32:27 now you sort of the weighted mean of haven deposits

32:30 to GDP where with weights coming from the share of aid

32:33 uh

32:34 in the total sample that these countries receive.

32:37 Then actually

32:38 this new weighted mean is significantly lower.

32:41 It's 0.014

32:43 and so the implied leakage rate would be 3.4 times 0.014

32:48 is roughly

32:50 5%.

32:51 So this is why we claim that sort of the aggregate leakage rate

32:54 is roughly

32:55 5%,

32:56 but of course,

32:57 There's a lot of uncertainty going into this calculation,

33:01 uh,

33:01 including statistical

33:02 uncertainty,

33:03 but I think

33:04 As a ballpark estimate,

33:05 probably we sort of get the order of magnitude.

33:08 Correct.

33:11 Uh.

33:12 Then

33:13 you may wonder,

33:14 OK,

33:14 why did you focus on these 22 highly a dependent countries?

33:18 Were you cherry picking?

33:20 So what we do in this graph is where we show how our point estimates,

33:23 which are depicted in blue,

33:25 change

33:26 when we change this threshold.

33:28 So when you move from the left to the right,

33:31 so we include progressively fewer,

33:33 fewer.

33:34 So,

33:35 have we done this analysis just for countries where World Bank aid exceeds 1% of GDP.

33:40 And actually we would have concluded that there is no significant correlation

33:44 between aid

33:45 and haven deposits.

33:47 But

33:48 as we move up,

33:50 we see that this coefficient estimate increases.

33:53 So in then we report

33:55 these estimates.

33:56 We we to go further,

33:58 we would typically find even higher.

34:01 Go fishing.

34:03 And

34:04 the implied leakage rate,

34:05 which is

34:07 depicted in red.

34:09 Also increases because these countries also

34:11 are made more aid dependent also seem to have an average sort of

34:15 higher ratios of having deposits to,

34:18 to GDP.

34:20 So

34:21 When looking at sort of this graph,

34:23 I think it's important to bear in mind that sort of

34:27 OK,

34:27 so yeah,

34:27 it's important to bear in mind that.

34:29 Particularly as you move towards the tail end of this distribution,

34:33 so we have fewer and fewer and fewer countries.

34:35 So,

34:37 and the data are relatively noisy,

34:38 so

34:39 perhaps we shouldn't make

34:40 too much of this,

34:41 but if you were asking me like what could be driving this pattern,

34:44 that sort of

34:45 the most intuitive explanation is that perhaps

34:48 Countries that sort of

34:50 receive the most aid are also countries where governance challenges are,

34:53 are greatest.

34:55 So,

34:56 now I want to conclude by sort of discussing

34:59 potential mechanisms.

35:01 So,

35:02 first I want to discuss some mechanisms that are almost surely not

35:05 driving what we see

35:07 in the data.

35:08 So

35:09 one explanation that some people

35:11 have sort of alluded to is that

35:13 oh this could be just central banks or

35:16 uh

35:17 regular banks

35:18 adjusting their portfolios.

35:19 Well,

35:20 those are not part of the data so

35:21 that really cannot be driving

35:24 what we're seeing.

35:25 The second explanation is that this could reflect profit shifting by

35:29 multinationals,

35:29 but we only observe

35:31 immediate

35:31 ownership.

35:33 Um,

35:33 so

35:34 this is also not part of our data.

35:36 Third,

35:37 you could say,

35:37 well,

35:37 this might just reflect general economic stimulus,

35:40 right?

35:40 So

35:41 aid tends to be good for economic activity,

35:44 um,

35:44 but in our estimates,

35:45 GDP growth is already

35:48 accounted for,

35:49 plus

35:50 the output response

35:51 of economies to

35:52 economic stimulus tends to be relatively more

35:55 protracted.

35:56 And this wouldn't explain why

35:59 money only flows

36:00 to havens

36:01 and not

36:02 to non-hans.

36:06 A second possible explanation is

36:08 elite capture.

36:10 So

36:10 this is definitely consistent with money flowing to havens but not to non-havens.

36:16 It could also rationalize why

36:18 aid inflows and capital outflows

36:21 concur

36:22 within

36:23 the same quarter.

36:25 And

36:28 To the extent you want to take sort of heterogeneity analysis seriously,

36:32 we,

36:32 we do find marginally higher

36:34 uh estimates

36:36 in countries with higher levels of corruption and,

36:38 and where disclosure requirements are present,

36:40 but I think we should,

36:41 you know,

36:42 take

36:42 those results with

36:44 uh

36:45 A grain of salt because.

36:48 Uh,

36:49 they're not statistically significantly different from another.

36:53 Then perhaps one of the most salient alternative explanation is that

36:57 this is a story about contractors executing,

37:00 uh,

37:00 a sponsored,

37:02 um,

37:06 Project.

37:08 And this is definitely consistent with sort of

37:09 finding slightly larger effects when the financial sector

37:13 is small,

37:13 so it could be the case that

37:15 Contractors prefer to use sort of foreign banking systems relative to their own

37:20 relatively underdeveloped

37:22 banking systems.

37:23 But again,

37:24 there's differences.

37:25 Not statistically

37:27 significant.

37:28 But

37:29 it's inconsistent with not finding larger effects when a

37:32 greater share of procurement goes to foreign firms,

37:35 and it also doesn't explain why money only flows to havens and not to non-hans,

37:39 which we've seen

37:42 are

37:43 what residents seem to be preferring

37:46 on average.

37:47 So

37:48 to conclude,

37:49 I hope that

37:50 uh

37:51 I've convinced you that this correlation exists,

37:53 that

37:54 World Bank

37:54 aid disbursements.

37:57 To coincide

37:59 with more rapid accumulation

38:00 of haven deposits.

38:03 But we only see this correlation in extremely aid dependent

38:07 countries.

38:08 It does not apply

38:10 across the board

38:11 and I would argue that these patterns are consistent with elite capture,

38:14 but

38:15 other explanations are

38:16 certainly possible.

38:34 All right.

38:51 All right,

38:52 uh,

38:53 thanks,

38:53 everybody.

38:54 Um,

38:56 So

38:59 I just wanna,

39:00 I know everybody's here to

39:02 um throw their hardest questions at Bob,

39:04 so I'll try to be brief,

39:05 uh,

39:06 and get out of the way so you can do that.

39:07 Um,

39:08 but

39:10 I just want to say,

39:11 and I don't know who needs to hear this,

39:12 but I think it's important to kind of kick off

39:14 by saying kind of the nature of the conversation that I think

39:17 we're having here and that we need to have,

39:18 which is that

39:19 the paper,

39:20 in my view,

39:21 um,

39:22 Rigorously establishes a new fact about the world,

39:25 and it's a fact

39:26 that we all need to kind of wrestle with.

39:29 And so most of my comments here are gonna be kind of taking,

39:31 I'll talk a bit about maybe a few concerns,

39:34 but taking Bob's core result at face value.

39:36 Um,

39:37 and there's lots of room for arguments about interpretation,

39:40 and that's kind of what I'm going to do,

39:42 but I think it's an extremely careful paper on an important topic

39:45 that to me

39:46 is quite persuasive,

39:48 um,

39:48 in its main contention here and that main kind of

39:51 fact about the world that I think we're all.

39:52 He wrestling with and as Bob pointed out,

39:54 media around the world are wrestling with what it means

39:57 is that aid disbursements to highly aid dependent countries

40:00 cause,

40:00 and I'm going to be less shy about the causal language

40:03 than the authors are,

40:04 um,

40:06 cause increased deposits

40:07 in tax havens.

40:09 And I think the strength of that analysis

40:11 comes off of,

40:12 you know,

40:12 the unique access to this BIS locational banking services data so they can see

40:17 who owns or from what country

40:20 the owners come for these deposits in different

40:22 tax havens and non-tax havens quarter by quarter.

40:26 I really I think commendable,

40:28 simple and transparent kind of differences and differences

40:31 frameworks so we're not lost in having to buy into a structural

40:34 model that we can't understand and you can

40:36 see exactly where the results are coming from,

40:39 and it's corroborated by this more flexible event study

40:41 um analysis.

40:43 And this IV estimation building on arts,

40:47 you know,

40:47 QJE paper

40:49 that addresses sort of our standard causality concerns,

40:52 you know,

40:52 is correlation causation.

40:54 You know,

40:55 we can never be 100% sure,

40:56 but this paper goes a long ways

40:58 towards establishing,

40:59 I think,

40:59 the causal chain

41:00 up there in,

41:01 in Blue Font.

41:03 Um,

41:03 the paper's full of myriad robustness checks,

41:06 um,

41:06 and Bob

41:07 walked us through several of those

41:09 already.

41:10 And I think kind of most importantly in the write-up and,

41:12 and certainly today in the presentation of the paper

41:14 is kind of a transparent recognition of what the paper can and can't show.

41:18 When you read the paper,

41:19 and most of what I'm gonna

41:21 comment on here in terms of

41:22 ah my questions of interpretation are things that are already noted

41:26 in the text of the paper,

41:27 and I would like to say that,

41:28 oh wow,

41:29 I thought of something clever and new,

41:30 but for the most part,

41:31 anything that I thought up,

41:32 I'm gonna say here,

41:33 you'll find on page 16 and,

41:35 and,

41:35 you know,

41:35 and so on.

41:37 Um,

41:38 so just 3 comments that I do want to focus on,

41:40 and I'm in the weird position

41:42 of

41:42 kind of standing here at the World Bank and trying to give you a more optimistic spin,

41:46 I think,

41:47 on some of the

41:48 results,

41:48 which for those of you who know me is not,

41:50 you know,

41:50 often the role,

41:51 um,

41:52 that I play,

41:53 but I've one comment on the sample,

41:54 the choice of sample,

41:56 one on kind of what the magnitudes mean and what we should make of this 5%

42:00 number,

42:00 um,

42:01 and then probably most significantly on the,

42:03 on the proposed mechanism that's advanced here.

42:06 Um,

42:07 this,

42:08 and I wasn't sure exactly what Bob was gonna show when I made these slides,

42:10 so this has kind of been

42:12 covered perfectly in Bob's presentation.

42:14 Um,

42:15 I think it is worth pointing out,

42:16 maybe mostly if we're,

42:17 you know,

42:18 discussing the paper in public,

42:19 is,

42:20 you know,

42:20 there's this question of,

42:21 is the average leakage of World Bank aid distinguishable from zero,

42:24 and I think the results

42:25 possibly imply that it's,

42:27 it's not.

42:27 So the focus here is on 22 highly aid dependent economies.

42:31 Um,

42:32 and I would have liked to see a little bit more justification.

42:34 I think I would have come in ex ante saying like,

42:37 oh,

42:37 I expect to see more aid in more corrupt places,

42:39 sorry,

42:40 more leakage in more corrupt contexts,

42:42 but I'm not sure I necessarily would have come in ex ante expecting to see

42:45 higher leakage rates

42:47 in more aid dependent.

42:49 economies,

42:50 um,

42:51 but that's what we find.

42:52 And so the paper focuses on those 22 countries where the effects are relatively

42:57 larger,

42:58 you know,

42:58 is that real heterogeneity or is it just easier for us to pick up

43:01 this leakage

43:03 when there's more signal to noise because

43:05 aid flows are a bigger share of what's going on in the economy?

43:07 I'm not sure.

43:09 But if we go to the lowest threshold,

43:10 as Bob already noted,

43:11 which becomes more representative and comprehensive of

43:14 the coverage of all World Bank lending,

43:16 then the results become indistinguishable

43:18 from zero.

43:19 And so it might be

43:21 reasonable to say

43:22 for overall bank lending,

43:23 the best evidence we have so far is that there's no statistical evidence

43:27 of leakage.

43:30 OK,

43:31 2,

43:32 on absolute versus relative magnitudes of the coefficient.

43:35 Um,

43:37 The absolute magnitude.

43:39 What do we make of this 5% number?

43:40 Well,

43:40 it's greater than 0,

43:41 and I think that's the most important thing.

43:43 That's,

43:43 you know,

43:44 that took a lot of work to get that.

43:45 So,

43:46 um,

43:47 that's an important point.

43:48 But the 5%,

43:49 I'm less convinced on how informative

43:51 that number is.

43:53 Um,

43:54 Again,

43:54 this is all on the paper.

43:55 That number is potentially too high if World Bank aid crowds in

43:59 other aid.

44:00 If a $50 million World Bank disbursement is matched

44:03 with a $50 million US aid or DFID disbursement.

44:06 Then the given leakage dollars

44:08 are compared to a bigger denominator and that leakage rate is smaller.

44:12 There's some counter-evidence provided in the paper,

44:14 but there's only so far you can go because

44:16 the other donors don't give us quarterly disbursement data.

44:19 Probably a bigger concern

44:21 is that this is an underestimate,

44:23 um,

44:24 because

44:25 without being able to see through shell corporations and

44:27 all the other ways this money is held,

44:29 um,

44:29 there could be a lot more movement to tax havens going on

44:32 than is captured

44:34 in this part of the BIS data,

44:36 in which case,

44:37 you know,

44:38 we could be looking at a bigger number.

44:39 So I don't know what exactly to make of the number and its absolute magnitude.

44:44 I think what's particularly interesting about the paper

44:46 is we can talk about relative magnitudes in a meaningful

44:51 and apples and apples way for once,

44:54 and there

44:55 we find that the leakage

44:56 is not significantly declining despite reforms.

44:59 I think that's news and kind of bad news.

45:03 I'm going to take statistical significance,

45:05 you know,

45:06 uh,

45:06 as a real filter here and say that it's

45:09 leakage is not significantly associated with corruption,

45:11 disclosure rules,

45:12 capital account openness,

45:13 democracy,

45:14 credit market debt.

45:15 Lots of the things we expected to predict this leakage rate

45:19 aren't predicting it significantly.

45:21 But maybe,

45:22 you know,

45:23 if I was working in World Bank

45:25 press,

45:26 um,

45:27 most notably is I could make a pitch that a big

45:29 finding of the paper is that aid is not oil.

45:33 And there's been this literature which has fretted for years and years.

45:37 Paul Collier,

45:37 you know,

45:38 literally a paper is aid oil.

45:40 Janko Vidal in 2008,

45:42 Steve Knack's papers

45:44 worried about aid as a sovereign rent,

45:46 which is going to have a

45:47 corrupting influence and undermine domestic institutions.

45:51 But what's nice about this is,

45:52 you know,

45:54 Bob can compare his results to earlier work by co-authors Anderson et al,

45:57 this 2017 GEO paper,

45:59 which is saying 15 to 20% of oil rents are flowing to these same tax havens.

46:04 One reading of the results would be,

46:06 wow,

46:06 the controls in place on bank disbursements are so good

46:10 that the majority of the leakage that's happening

46:12 through other transfers to developing countries are not happening

46:15 with World Bank loans.

46:16 Again,

46:17 glass 2/3 full,

46:18 but it is

46:19 another reading of the same result.

46:23 Finally,

46:23 on the proposed

46:25 mechanism,

46:26 um,

46:28 The proposed mechanism in the paper

46:30 is that this comes from corruption and

46:31 embezzlement by ruling politicians and bureaucrats.

46:34 I agree with Bob that I think this is a plausible mechanism.

46:37 Maybe if I had to put money on,

46:39 you know,

46:39 we were going to find out with the god's eye view what it was.

46:42 Maybe that's where I put my money,

46:44 but I do think we've got to be

46:46 fairly open to alternative.

46:48 Mechanisms and here I'm mostly kind of serving as

46:51 rapporteur on the active Twitter conversation going on

46:55 around this paper,

46:56 um,

46:56 including lots of bank veterans chiming in

46:59 with their experience about what could be going on here.

47:01 Um,

47:02 one being aid disbursements leading to currency appreciation,

47:05 triggering capital flight.

47:07 Um,

47:07 disbursements to Ethiopia,

47:08 I mean the bur appreciates and if you were in Ethiopia looking to get money out,

47:12 now's a good time,

47:13 um,

47:14 when the bur is strong.

47:16 Um,

47:18 Alternatively,

47:19 to avoid appreciation,

47:21 for some people suggest that policymakers might

47:23 actually loosen capital controls momentarily and allow

47:26 Forex to escape

47:28 in the window around the World Bank disbursement.

47:31 So again,

47:32 people looking to move money overseas,

47:33 taking advantage of this window

47:35 right around the time of disbursement to do so.

47:39 Um,

47:40 and 3rd.

47:42 Something that Bob referred to as well,

47:43 that vendors choose to hold

47:45 potentially illicit and legal

47:47 revenues offshore

47:49 for for possibly illicit reasons.

47:52 I think

47:53 a key to all of these alternative mechanisms is

47:56 there's still big news here.

47:58 We still have a tax havens problem

48:00 where people are motivated to put money in tax havens,

48:03 quite possibly for tax evasion reasons,

48:06 but these stories don't necessarily begin with corruption and diversion

48:10 of World Bank monies.

48:12 Hat tip to all the former World Bank economists on Twitter,

48:15 um.

48:18 And just kind of continuing on that point,

48:20 I'm

48:21 probably.

48:24 Least solved by

48:25 what the empirics can give us

48:27 and what continued regressions can give us

48:29 to distinguish between

48:31 these different

48:33 mechanisms.

48:34 So,

48:35 for the corruption and embezzlement mechanism,

48:37 the paper offers corroborating evidence,

48:38 you know,

48:39 that there are bigger effects in more corrupt countries.

48:42 Yes,

48:42 but not significantly so,

48:45 um,

48:45 that the timing of the effect is,

48:46 you know,

48:47 T equals 0,

48:49 which would be

48:50 consistent with a politician skimming off of the transfer,

48:53 but could be

48:54 consistent with alternative stories.

48:56 I'd argue

48:57 the effect is only present for tax havens.

49:00 I think that's

49:01 a big important part of the paper,

49:03 um.

49:04 But here I want to be particularly careful because I think there's an

49:07 established assumption in this literature,

49:09 like in the Zuman,

49:11 Gabriel Zucman's work,

49:12 to say that deposits in tax havens

49:15 are sort of prima facie evidence of tax evasion.

49:18 That's one step,

49:19 but that's,

49:20 it's a big step further to say that deposits in

49:23 tax havens are evidence that those were ill-gotten gains.

49:26 That's a

49:26 different assertion,

49:27 and I don't think the previous tax havens literature has made that assertion,

49:31 assertion.

49:32 So I'm not sure we can use this

49:34 as evidence that the gains themselves were ill-gotten.

49:37 Um,

49:38 and then.

49:40 You mentioned different people had different priors here.

49:42 I guess my prior would have been

49:44 that if this was skimming off of World Bank transfers,

49:46 I would have expected to see bigger

49:48 effects for the DPLs than for the investment projects.

49:51 That's

49:52 More fungible money,

49:54 um,

49:54 and we,

49:55 we just don't

49:56 see that here.

49:58 Um,

49:59 To wrap up,

50:00 I mean,

50:01 I think rather than trying to tease out

50:04 more of this story with heterogeneous effects,

50:07 at some points the samples,

50:08 you know,

50:09 we're talking about a fairly small number of countries,

50:11 and I'd be really interested to see

50:13 some triangulation of the

50:15 great econometric evidence you have on the core result,

50:18 with some qualitative support about can you find,

50:21 Evidence,

50:22 uh,

50:22 you know,

50:22 we have,

50:23 I was asking former bankers,

50:25 you know,

50:25 the Padma Bridge scandal in Bangladesh or the Lesotho Highlands Water project,

50:29 like,

50:30 do the known cases of diversion of bank monies

50:33 fit into the mechanism that you're describing,

50:36 and can you do some sort of like,

50:38 you know,

50:38 you can look for influential observations,

50:40 we can do the kind of aeronau and Sammy stuff to say,

50:43 who are the observations that are driving the results.

50:46 And then I don't know,

50:47 send it out to the 1818

50:49 H Street Club,

50:50 what is it called,

50:50 and say,

50:51 you know,

50:52 do people have stories about those influential observations?

50:54 But I,

50:55 I think a little more qualitative work

50:57 to back up that mechanism would make the paper a lot,

50:59 would be fun to read,

51:00 um,

51:01 and would make me feel better,

51:02 um,

51:03 about this uh

51:04 corruption mechanism.

51:05 So in the end,

51:06 On the sample,

51:07 if I was going to be super optimistic here,

51:09 average leakage of World Bank aid is indistinguishable from zero.

51:12 On the magnitudes,

51:13 the main takeaway is,

51:14 hey,

51:15 aid's a lot better than oil in terms of our leakage rate,

51:17 and on the proposed mechanism,

51:19 are we 100% sure that this is really corruption and not something else

51:22 going on?

51:24 Let me stop there,

51:25 but overall,

51:26 really fun paper,

51:27 really persuasive main result.

51:28 Thanks a lot.

showAllTimestamps
no
transcript
Um, welcome, everyone. Thanks for coming on, uh, on a Friday afternoon to, uh, uh, to hear a very interesting paper, Elite Capture and Foreign Aid by our colleague Bob Rikers, who is a, uh, senior economist in the Development Research Group in DEC in the World Bank, co-authored with Jurgen Andersson and Niels Johansson. Um, you know, this is a paper that's been in the news a lot lately, and we all know that it's perilous to get information from headlines, and so we thought that it would be very useful to have a seminar in which, uh, Bob could actually present the entire paper as opposed to sort of the, the, the headlines that have been making the rounds in, uh, in, in Twitter and in the news and so on. I, I wanted to say, uh, just one other quick thing about the paper. So, The papers, uh, in the World Bank's working paper series and it has, uh, garnered 21,000, 20, you're right, 21,000 downloads, um, since last Tuesday when it was posted. This makes it the, uh, fourth most downloaded World Bank working paper, um, uh, sort of in terms of lifetime downloads since 2014 when we started tracking it. Now you might wonder what the most downloaded World Bank working paper is. And for those of you who doubt that the World Bank has a well-developed sense of both humor and irony, the most downloaded paper is a paper by our colleagues Dorta Dumelund and James Trevino, which is called Which World Bank Reports Are Most Widely Read. Uh, I, I'm not making this up. That is actually, so, anyhow, so this all fits very nicely together and it's a, it's a great pleasure to have, um, Bob here to present, uh, um, uh, to present about the paper. Um, because it's a large room, we won't do our regular sort of seminar format of questions along the way, but rather we'll hear first from Bob, then we'll hear from our discussion, uh, Justin Sandifer. Justin is the co-director of Education Policy and senior fellow at Center for Global Development, which is, despite being a broad title, extremely narrow compared to what Justin's. Interests are and what his influence is in the developmental policy discussions. He's worked extensively also on aid effectiveness, which makes him a particularly apt, uh, discussant for this talk, and he's really been, uh, a valued colleague in the Greater Washington area who I always enjoy talking with and having him criticize my research. So, hopefully you'll do Bob the same courtesy. And, uh, so Bob will talk for about 40 minutes, maybe just in 10, and then we'll open up for, uh, uh, for Q&A. So thank you again and over to you, Bob. Oh, OK. Good afternoon. Thanks very, very much for coming and, uh, for having me. I'm here to present my paper, Elite Capture of Foreign Aid Evidence from Offshore Bank Accounts, which is joint work with Niels Johannesen and Jurgen Juel Andersen. But I would also like to use the opportunity to thank uh our research assistant Nicholas Gomez-Para. I'm not sure whether, where he is there, like, he's excellent and he's really made like big contributions. I also want to use the opportunity to thank many of you uh for providing comments. If you feel that we haven't addressed your comment, it's not because we're not taking it seriously, it's probably because somebody else made a comment suggesting to do the exact opposite. So, uh, with that, um, I would like to start. So, a question that concerns Certainly everybody in this room, but I would argue the entire development community is does aid reach its intended beneficiaries? And the concern is that it may not, that some part of it may be captured by those who need it least, the elites. And those concerns are certainly consistent with economic theories of rent seeking. And empirical evidence suggesting That aid doesn't necessarily flow to countries that are less corrupt. And this in some sense is pretty intuitive because if countries were amazingly well governed, perhaps they wouldn't need so much aid in the first place. But of course, this phenomenon is very difficult to measure. And this is typical in the literature on corruption. One often has to rely on very indirect evidence, and this paper is no exception. So the basic question. We ask in this paper. To address this concern is do World Bank aid disbursements trigger increases in foreign deposits in bank accounts, he in havens? The So what are havens? Havens are offshore banking centers known for secrecy and asset protection. Such as Switzerland, Luxembourg, the Cayman Islands. And our main sample is going to comprise 22 countries. That are really reliant on foreign aid, but World Bank aid alone. Accounts for more than 2% of GDP. Each year over our sample period, which is from 1990 until 2010, and our overall aid. is on average 10% of GDP. But for context, I want to emphasize from the beginning that these countries actually only absorb approximately 10% of all World Bank lending. And an extensions of the paper, I'm going to show you sort of what has happened in the period since 2009 when there's been a significant crackdown on, on tax havens. And what our results are for less aid dependent countries, and I'm going to show you that in those countries we do not see this pattern. Um, that we see in the most independent countries. So the premise of the paper is that haven deposits belong to elites. How do we know this? Well, evidence from Scandinavia suggests that the top 0.01% richest households. Own 50% of the wealth held in havens and the richest 0.1%. Own 80% of that wealth. And so there are similar studies in Colombia that sort of confirm. That this offshore world is extremely concentrated at the top of the distribution. By contrast, The poor are typically unbanked. So thanks to the work of Ashley and our colleagues, uh, in the finance group and DC. We know that in our sample of 22 aid dependent countries in 2011, Among the bottom 40% poorest households, only 9% had a domestic bank account, so it's entirely implausible, I would argue that the poor are controlling the money flowing to havens. So that's the premise. So what's our main finding? Our main finding is that World Bank aid disbursements. Do these highly aid dependent countries coincide with significant increases. And haven deposits within the same quarter. And How, how big is this increase? Well, We define something we call the leakage rate, which is the aid induced increase in haven deposit. And that Equals roughly 75% of aid inflows for the average country. But because most of the aid actually flows to, to countries where the stock of having deposits to GDP is relatively low. As a share of all aid going. To these 22 countries, this number is roughly 5% and intriguingly, This leakage rate tends to increase with the ratio of aid to GDP as I will show you later. By contrast, We do not find any evidence that a disbursements trigger flows to non-havens. Like New York. Or London. Or Gothenburg. And there's also no evidence that this correlation we document between aid disbursements and the accumulation of haven deposits has changed over the last decade. So I hope that by the end of the talk, I'll, I'll, I've managed to convince you that sort of this, this correlation exists and is very real. What I think is, is much harder to figure out is what is driving this correlation that we document. And that's because we simply don't know who owns these deposits because we have macrodata and because we cannot track individual financial flows, it's very difficult to attribute agency. Nonetheless, I'm going to argue that elite capture is an explanation that's certainly consistent with the totality of the patterns we observe, but Other explanations are possible with our data, unfortunately, you're not going to be able to prove or disprove any of them. And if you've any ideas or suggestions, I would very, very, very much welcome them. Uh, but before proceeding, I think it's important to highlight some of the important caveats. Uh, of our analysis and you know some misperceptions that, that I think have made headlines. So for instance, for instance, according to the Times, Billions of foreign aid disappears into tax havens. Yes, quite possibly based on our analysis, but correlation doesn't necessarily imply causation, so we cannot be confident that this is what is driving our results. And on top of that, money is fungible, so it's not clear that if diversion is taking place, that diversion is out of aid flows as opposed to, say, money that was earmarked for the Ministry of Education. Second, some newspapers have, have sort of made what I deem relatively grandiose claims about sort of the, the magnitude of leakage we document. So for instance, according to Forbes, as much as like 15% of foreign aid flows into uh tax havens. Uh, I think this is a wild exaggeration because again, Our main south pole comprises 22 countries that together absorb 10% of all World Bank lending. We find a leakage rate of roughly 5%. We don't find. Leakage sort of in less aid dependent countries, so expressed as a share of the World Bank's overall portfolio, the true number is much likely to be closer to 0.5%. So, so it's not the case that sort of this pattern applies across the board, which I think is very important to bear in mind when we discuss our results. So this concludes the introduction of the paper. Next I'll tell you a little bit about our data, our empirical strategy, uh, the various robustness tests we've done, and I would like to spend more time on the interpretation and this is also something where, you know, I would welcome your, your, uh, feedback because many of you have, have expertise that I think is extremely relevant to interpreting these findings and then I'll wrap up. So our main data set is a data set of cross-border bank deposits, sorry, non-bank deposits, so it excludes financial institutions, banks, including the central bank. Uh, so you can think of these as privately held account firms and individuals, but we only observe immediate ownership of these. This data is confidential and comes from the locational banking statistics of the Bank for International Settlements, which covers, uh, bank deposits and roughly 43 financial centers owned by residents of around 200 countries. And so we use confidential data that covers all these uh financial centers for the period 1919 to 2010. But then we also replicate our analysis with the publicly available BAS data which allows us to also examine what happened over the last decade. But these data have much more partial coverage because they're based on sort of voluntary reporting of, of countries. Um, and our key dependent variable is going to be the sum of all haven deposits owned by residents of a particular country in the 17 havens in a particular quarter. So think about sort of All the money that residents of Kyrgyzstan have in Switzerland, that they have in Luxembourg, that they have in Belgium, like any of these 17 havens, and they were going to add these up. For foreign aid, we use information on project level disbursements from the World Bank project database. Now there's a real irony in this work because the reason we focus on the World Bank is that it is in fact the only major donor for which we can construct a series at the quarterly level for a long period of time. So because of the transparency, In our institution, I think we're now under scrutiny. Um, so, so in this data, we know for each project, project when it was approved, how much money was committed, what sector, uh, The project was in, the type of lending instrument and the timing of each disbursement, and this is going to be extremely useful when we try to Address potential endogeneity of it because we can use this to predict. How much aid we would expect to flow to these countries if Aid would flow sort of typical patterns. So, Our measure of aid is simply the sum of all these project level disbursements on iodine IBRD projects in a given country in a given quarter. So to be clear, this excludes debt relief and it also excludes uh trust funds. In addition, we, we, uh, use a number of, uh, other data sets, so we're going to exclude periods of war, coups, crisis, disasters, uh, we look at how the results vary with petroleum, rents, financial sector development, the capital account, and disclosure requirements for politicians. Uh, you're also gonna look how it varies with democracy, uh, corruption, and importantly with sort of the share of aid that is procured domestically versus internationally. Because you might think that sort of these patterns that we're documenting are just an artifact of, of contractors who execute a sponsored contracts. Depositing some of their profits in in havens, and I'm going to argue that, yeah, maybe that's, maybe that's happening, but it's quite unlikely. So, Just to give you a sense that these are distributions of aid as a share of GDP and, and even deposits growth. As you can see, there's quite a lot of variation in in aid inflows, which is very useful for identification. And you also see, like, when you look at the distribution of haven deposits growth, that in many quarters, they actually go down. So, so there is this saving, which I think is relevant to bear in mind. So, and some descriptive statistics on, on, on the sample. So World Bank aid in the sample accounts for roughly 2.7% of GDP on average. Overall aid, uh, It's roughly 10% of GDP. So these are economies in which aid is really a big deal. Uh, when we look at the level of haven deposits, we see that It stands on average around 61 million and the level of haven deposits stands around 136 million. So this I think is an extremely relevant descriptive statistic because it suggests that On average, It would seem that residents of these countries have a strong preference for depositing money in non-hans relative to havens, so that makes it all the more puzzling that when aid flows in, we only see outflows. To havens and not to non-hans, so it's not simply not true that, you know, residents of these countries routinely Prefer depositing money in havens. Uh So, our empirical strategy is extremely simple. We're going to regress the growth in haven deposits measured in logs on aid as a share of GDP. Controlling for GDP growth and in robustness checks, we also had controls for Um, the mechanical appreciation in deposits that would result from, from exchange rate movements alone. As well as commodity price shocks because you might expect that sort of some of the wealth that's going into havens is the result of these commodity price, price movements. Uh We're going to control also for country and time fixed effects. So effectively we're asking whether a country when it receives aid. Accumulates deposits more rapidly than in times when it doesn't. Uh, standard errors are going to be clustered at the country level, which is conventional, and we're going to Windsorize our deposit and aid measures, uh, at the 1% level because there are some extreme outliers, but these are this Windsorization is not driving the results as I'm gonna. show you. And then in alternative specifications, we also look at the growth in non-haven deposits and the differential growth of haven deposits relative to non-haven deposits. So if you thought that, you know, this was just a story about sort of Aid sort of stimulating economic performance and leading to capital outflows, that wouldn't necessarily explain why there's differential, uh, growth outflows to havens versus non-havens. But of course, a major problem here is aid is deeply endogenous. So it's quite likely that, you know, a country receives aid at a time when it needs it most, for instance, when it's experiencing A natural disaster or a financial crisis, when I would argue it's also particularly appealing to take your money out of the country and park it somewhere else. Um So We're going to address this in 180 in three ways. First, we're just going to augment our specification with both lags and leads of our aid variable to make sure that it's really aid that is driving the surge in haven deposits so that there are no pre-trends. Secondly, we're going to use an instrumental variable strategy, drawing on an excellent paper by Art who had a simple but powerful insight that how much aid a country receives in any given quarter is to a very large extent determined by past commitments that cannot respond to sort of contemporaneous events, contemporaneous shocks. And so if we take those past commitments and sort of we apply sort of the typical disbursement schedule that applies to World Bank loans. Then we can come up with sort of a synthetic measure of predicted a disbursements that's plausibly exogenous that we can use as an instrument. And the third, we're going to exclude the episodes of turmoil. Um, such as war, coups, uh, conflict, and financial crisis. So, these graphs show you the first stage of our instrument with aid residualized on the left hand side and predicted aid on the right hand side. And as you can see, the correlation is, is really strong. So, as I already sort of explained, it's based on sort of predicting project level quarterly disbursements based on sort of the typical disbursement profile for Projects in the same sector, in the same region, executed in the same time period. And then simply aggregated those predicted disbursements to get a measure of sort of country level predicted World Bank aid. Um So on to the main result, so. This table shows you our, our main results, so. It shows that When It increases. The growth in haven deposits accelerates. In fact, if aid goes up by 1%, the growth of in haven deposits accelerates by roughly 3.4%, and we do not see a similar response of non-haven deposits. So the differential which is presented in the 3rd column. Between haven versus non-haven deposits growth is also statistically significant. I would also like to use the opportunity to point out that when we look at the coefficients on GDP growth, those are much more similar for both haven and non-haven deposits growth. So they're 0.12 and 0.15. So it seems that both haven deposits and non-haven deposits responded relatively. Symmetric proportional fashion to, to sort of general increases and, and or decreases in economic um activity. So sort of this, this asymmetric response of haven deposits seems to be very particular to, to aid. But you may be concerned that perhaps sort of there were earlier events that triggered both uh aid inflows and, and the accumulation of having deposits. So what this graph here does, it sort of shows you. Um, the results of a specification in which we add lags and lead. And so the quarter relative, the, the Dots are sort of point estimates for the coefficient estimates of the lags and leads and the, the bars are the confidence intervals. And as we can clearly see, sort of we see a big spike. On At the time the aid arrives, but no evidence sort of of responses to aid received in earlier quarters or to aid that's about to, to come in. Uh, this is the same graph for non-haven deposits which don't seem to significantly respond to aid inflows, although they're in a small, insignificant dip, uh. At times 0 and of course then the difference. Between the accumulation of haven deposits and non-haven deposits is also Yeah. Exhibiting a very similar pattern as the one we saw two slides ago. Yeah And how robust is this result? Well, when we instrument aid, What we see, which we do here in columns 2 and 3, where the difference between the 2nd and the 3rd column is that we exclude sort of more of the initial quarters to, to have sort of uh ar more exogeneity. And you see that sort of the coefficient drops somewhat, as you would expect, but it remains. Similar in magnitude, I would argue, and Strongly statistically. Significant. So this should sort of assuage a lot of the concerns about endogeneity. Nonetheless, we also examine what happens when we exclude periods of wars, coups, disasters, and financial crisis, uh, and you can see that our main finding. holes in the coefficient remains similar in size and also strongly statistically significant. And it's also true when we sort of exclude all of these at the same time, which I'm not presenting here. Then we also run a number of alternative specifications. So crucially, in column 5, we include country year fixed effects. So this is an extremely demanding test, I would argue, because now all the variation. is coming from looking at what happens to that same country. Uh In the same year, Uh And as expected, sort of the results are slightly less significant. They remain significant to the 10% level, but the coefficient estimate is very, very similar. In column 6, we include exchange rate control. So some people have said, oh, perhaps this is just something to do with sort of how exchange rates move. Well, here we sort of control for the mechanical appreciation of deposits that would eventuate because of exchange rate movements alone. This does not seem to meaningfully alter our estimated coefficient. Then in column 7 we control for resource dependence and fluctuations in the price of oil. Uh By interacting A dummy for being a resource, uh, being an oil producer with, uh, our time dummies, and the reason for doing this is that my co-authors have a similar paper where they examine what happens to resource windfalls and they show that in autocratic countries roughly 15% of um oil wealth seems to be flowing into tax havens but controlling sort of for this doesn't really Impact, uh, our estimates of the impact of aid. And then finally, in the final column, we show what happens when we just use the raw data and don't wins our eyes and as you can see, the results are strikingly similar to uh our main results. So I hope that by now you're somewhat convinced that sort of this pattern is real, but to understand the mechanism, I think it's important sort of to look at how this pattern varies across countries. Now, it's important to bear in mind that our sample is extremely small. We only have 22 countries, so naturally we don't expect sort of Huge differences, but nonetheless sort of to see whether at least sort of qualitatively these, these parents could be suggestive. What we do is sort of we split the sample. Every time into countries that sort of have above and below the median. Control of corruption, for instance. So here we see that. Outflow is slightly larger, but certainly not significantly larger in countries where the control of corruption is lower. We also see that outflows are slightly larger, but again not significantly larger in countries that require their politicians to disclose their assets, i.e., countries where You could argue politicians should have an incentive to, to use havens. The fact is also slightly higher in countries that are less democratic. Uh, and In countries With better developed financial systems. And last but not least, The effect is slightly higher in countries. That Procure more of They um. A larger share of the 8 contracts from foreign firms. So if you had expected that sort of this was a story about sort of domestic contractors depositing. They're legitimately earn profits in havens. You might not have expected. The, the opposite and that's not what we see. Now you could argue that, well, you know, these contractors must subcontract, so you know, whether you give it to a foreign firm or uh a domestic firm in the end, like they're all going to rely on the same domestic subcontracts that I think is like a legitimate contract critique sort of to, to this point. So, so I don't think like, you know, anything I'm showing is sort of uh conclusive, but I do think it, it raises questions. And third, I want to look at some of the type of aid, so. Uh Based on a lot of conversations sort of with country economists, my prior had been that if we believe that these patterns are the product of diversion, we should perhaps expect to see larger effects for development policy lending, where there's relatively less oversight, and that's actually not what we see in the data. We see, if anything, slightly stronger effects for projects. But then, uh, a colleague in the integrity vice president, he suggested that this is exactly precisely what he would have expected because his argument was that, well, most of the money is actually going through projects and if you want to capture these funds, you typically do it through politically connected firms. Now, whether that's a, uh, plausible argument, I think it's a matter of, of debate, but I frankly do not have a completely compelling explanation as to why we see these differences, but I should emphasize that these differences are extremely small and not Statistically significant, so I think it's also important not to overinterpret them. Uh. So then we replicate our analysis using publicly available data to see whether the crackdown on tax havens that we've been seeing in the past 10 years since 2009. has sort of had an impact on the patterns we see in the data. And basically the coefficient estimates we get for The period from 2010 to 2019 relative sort to a free period. Coefficient estimate is very strikingly similar in magnitude. But as you can see like the last period like has a much wider confidence interval so you know I don't think we can conclusively conclude sort of that transparency hasn't, uh, helped. But The point estimates suggest that the change has been very limited. OK, so how big is sort of the aid induced increase in offshore deposits, what we refer to as leakage? Well, this is not something our model uh delivers directly. So we have to rely on a simple transformation. So the coefficient we get on aid is 3.4. So it means that if aid goes up by 1%, growth in haven deposits accelerates by 3.4%. On average, The stock of haven deposits express as a share of GDP in our sample is 0.22. So when we multiply 3.4 with 0.22, we get 7.48 or let's say roughly 7.5% uh. Is the implied leakage leakage rates because if aid goes up by $1 haven deposits should go up at the sample mean by $7.5. But this Calculation applies to the average country, but it happens to be the case that most of the aid is actually flowing to countries with a relatively low stock of haven deposits to GDP. So if we do the same calculation using the same 3.4% number applying, assuming it applies uniformly. But now you sort of the weighted mean of haven deposits to GDP where with weights coming from the share of aid uh in the total sample that these countries receive. Then actually this new weighted mean is significantly lower. It's 0.014 and so the implied leakage rate would be 3.4 times 0.014 is roughly 5%. So this is why we claim that sort of the aggregate leakage rate is roughly 5%, but of course, There's a lot of uncertainty going into this calculation, uh, including statistical uncertainty, but I think As a ballpark estimate, probably we sort of get the order of magnitude. Correct. Uh. Then you may wonder, OK, why did you focus on these 22 highly a dependent countries? Were you cherry picking? So what we do in this graph is where we show how our point estimates, which are depicted in blue, change when we change this threshold. So when you move from the left to the right, so we include progressively fewer, fewer. So, have we done this analysis just for countries where World Bank aid exceeds 1% of GDP. And actually we would have concluded that there is no significant correlation between aid and haven deposits. But as we move up, we see that this coefficient estimate increases. So in then we report these estimates. We we to go further, we would typically find even higher. Go fishing. And the implied leakage rate, which is depicted in red. Also increases because these countries also are made more aid dependent also seem to have an average sort of higher ratios of having deposits to, to GDP. So When looking at sort of this graph, I think it's important to bear in mind that sort of OK, so yeah, it's important to bear in mind that. Particularly as you move towards the tail end of this distribution, so we have fewer and fewer and fewer countries. So, and the data are relatively noisy, so perhaps we shouldn't make too much of this, but if you were asking me like what could be driving this pattern, that sort of the most intuitive explanation is that perhaps Countries that sort of receive the most aid are also countries where governance challenges are, are greatest. So, now I want to conclude by sort of discussing potential mechanisms. So, first I want to discuss some mechanisms that are almost surely not driving what we see in the data. So one explanation that some people have sort of alluded to is that oh this could be just central banks or uh regular banks adjusting their portfolios. Well, those are not part of the data so that really cannot be driving what we're seeing. The second explanation is that this could reflect profit shifting by multinationals, but we only observe immediate ownership. Um, so this is also not part of our data. Third, you could say, well, this might just reflect general economic stimulus, right? So aid tends to be good for economic activity, um, but in our estimates, GDP growth is already accounted for, plus the output response of economies to economic stimulus tends to be relatively more protracted. And this wouldn't explain why money only flows to havens and not to non-hans. A second possible explanation is elite capture. So this is definitely consistent with money flowing to havens but not to non-havens. It could also rationalize why aid inflows and capital outflows concur within the same quarter. And To the extent you want to take sort of heterogeneity analysis seriously, we, we do find marginally higher uh estimates in countries with higher levels of corruption and, and where disclosure requirements are present, but I think we should, you know, take those results with uh A grain of salt because. Uh, they're not statistically significantly different from another. Then perhaps one of the most salient alternative explanation is that this is a story about contractors executing, uh, a sponsored, um, Project. And this is definitely consistent with sort of finding slightly larger effects when the financial sector is small, so it could be the case that Contractors prefer to use sort of foreign banking systems relative to their own relatively underdeveloped banking systems. But again, there's differences. Not statistically significant. But it's inconsistent with not finding larger effects when a greater share of procurement goes to foreign firms, and it also doesn't explain why money only flows to havens and not to non-hans, which we've seen are what residents seem to be preferring on average. So to conclude, I hope that uh I've convinced you that this correlation exists, that World Bank aid disbursements. To coincide with more rapid accumulation of haven deposits. But we only see this correlation in extremely aid dependent countries. It does not apply across the board and I would argue that these patterns are consistent with elite capture, but other explanations are certainly possible. All right. All right, uh, thanks, everybody. Um, So I just wanna, I know everybody's here to um throw their hardest questions at Bob, so I'll try to be brief, uh, and get out of the way so you can do that. Um, but I just want to say, and I don't know who needs to hear this, but I think it's important to kind of kick off by saying kind of the nature of the conversation that I think we're having here and that we need to have, which is that the paper, in my view, um, Rigorously establishes a new fact about the world, and it's a fact that we all need to kind of wrestle with. And so most of my comments here are gonna be kind of taking, I'll talk a bit about maybe a few concerns, but taking Bob's core result at face value. Um, and there's lots of room for arguments about interpretation, and that's kind of what I'm going to do, but I think it's an extremely careful paper on an important topic that to me is quite persuasive, um, in its main contention here and that main kind of fact about the world that I think we're all. He wrestling with and as Bob pointed out, media around the world are wrestling with what it means is that aid disbursements to highly aid dependent countries cause, and I'm going to be less shy about the causal language than the authors are, um, cause increased deposits in tax havens. And I think the strength of that analysis comes off of, you know, the unique access to this BIS locational banking services data so they can see who owns or from what country the owners come for these deposits in different tax havens and non-tax havens quarter by quarter. I really I think commendable, simple and transparent kind of differences and differences frameworks so we're not lost in having to buy into a structural model that we can't understand and you can see exactly where the results are coming from, and it's corroborated by this more flexible event study um analysis. And this IV estimation building on arts, you know, QJE paper that addresses sort of our standard causality concerns, you know, is correlation causation. You know, we can never be 100% sure, but this paper goes a long ways towards establishing, I think, the causal chain up there in, in Blue Font. Um, the paper's full of myriad robustness checks, um, and Bob walked us through several of those already. And I think kind of most importantly in the write-up and, and certainly today in the presentation of the paper is kind of a transparent recognition of what the paper can and can't show. When you read the paper, and most of what I'm gonna comment on here in terms of ah my questions of interpretation are things that are already noted in the text of the paper, and I would like to say that, oh wow, I thought of something clever and new, but for the most part, anything that I thought up, I'm gonna say here, you'll find on page 16 and, and, you know, and so on. Um, so just 3 comments that I do want to focus on, and I'm in the weird position of kind of standing here at the World Bank and trying to give you a more optimistic spin, I think, on some of the results, which for those of you who know me is not, you know, often the role, um, that I play, but I've one comment on the sample, the choice of sample, one on kind of what the magnitudes mean and what we should make of this 5% number, um, and then probably most significantly on the, on the proposed mechanism that's advanced here. Um, this, and I wasn't sure exactly what Bob was gonna show when I made these slides, so this has kind of been covered perfectly in Bob's presentation. Um, I think it is worth pointing out, maybe mostly if we're, you know, discussing the paper in public, is, you know, there's this question of, is the average leakage of World Bank aid distinguishable from zero, and I think the results possibly imply that it's, it's not. So the focus here is on 22 highly aid dependent economies. Um, and I would have liked to see a little bit more justification. I think I would have come in ex ante saying like, oh, I expect to see more aid in more corrupt places, sorry, more leakage in more corrupt contexts, but I'm not sure I necessarily would have come in ex ante expecting to see higher leakage rates in more aid dependent. economies, um, but that's what we find. And so the paper focuses on those 22 countries where the effects are relatively larger, you know, is that real heterogeneity or is it just easier for us to pick up this leakage when there's more signal to noise because aid flows are a bigger share of what's going on in the economy? I'm not sure. But if we go to the lowest threshold, as Bob already noted, which becomes more representative and comprehensive of the coverage of all World Bank lending, then the results become indistinguishable from zero. And so it might be reasonable to say for overall bank lending, the best evidence we have so far is that there's no statistical evidence of leakage. OK, 2, on absolute versus relative magnitudes of the coefficient. Um, The absolute magnitude. What do we make of this 5% number? Well, it's greater than 0, and I think that's the most important thing. That's, you know, that took a lot of work to get that. So, um, that's an important point. But the 5%, I'm less convinced on how informative that number is. Um, Again, this is all on the paper. That number is potentially too high if World Bank aid crowds in other aid. If a $50 million World Bank disbursement is matched with a $50 million US aid or DFID disbursement. Then the given leakage dollars are compared to a bigger denominator and that leakage rate is smaller. There's some counter-evidence provided in the paper, but there's only so far you can go because the other donors don't give us quarterly disbursement data. Probably a bigger concern is that this is an underestimate, um, because without being able to see through shell corporations and all the other ways this money is held, um, there could be a lot more movement to tax havens going on than is captured in this part of the BIS data, in which case, you know, we could be looking at a bigger number. So I don't know what exactly to make of the number and its absolute magnitude. I think what's particularly interesting about the paper is we can talk about relative magnitudes in a meaningful and apples and apples way for once, and there we find that the leakage is not significantly declining despite reforms. I think that's news and kind of bad news. I'm going to take statistical significance, you know, uh, as a real filter here and say that it's leakage is not significantly associated with corruption, disclosure rules, capital account openness, democracy, credit market debt. Lots of the things we expected to predict this leakage rate aren't predicting it significantly. But maybe, you know, if I was working in World Bank press, um, most notably is I could make a pitch that a big finding of the paper is that aid is not oil. And there's been this literature which has fretted for years and years. Paul Collier, you know, literally a paper is aid oil. Janko Vidal in 2008, Steve Knack's papers worried about aid as a sovereign rent, which is going to have a corrupting influence and undermine domestic institutions. But what's nice about this is, you know, Bob can compare his results to earlier work by co-authors Anderson et al, this 2017 GEO paper, which is saying 15 to 20% of oil rents are flowing to these same tax havens. One reading of the results would be, wow, the controls in place on bank disbursements are so good that the majority of the leakage that's happening through other transfers to developing countries are not happening with World Bank loans. Again, glass 2/3 full, but it is another reading of the same result. Finally, on the proposed mechanism, um, The proposed mechanism in the paper is that this comes from corruption and embezzlement by ruling politicians and bureaucrats. I agree with Bob that I think this is a plausible mechanism. Maybe if I had to put money on, you know, we were going to find out with the god's eye view what it was. Maybe that's where I put my money, but I do think we've got to be fairly open to alternative. Mechanisms and here I'm mostly kind of serving as rapporteur on the active Twitter conversation going on around this paper, um, including lots of bank veterans chiming in with their experience about what could be going on here. Um, one being aid disbursements leading to currency appreciation, triggering capital flight. Um, disbursements to Ethiopia, I mean the bur appreciates and if you were in Ethiopia looking to get money out, now's a good time, um, when the bur is strong. Um, Alternatively, to avoid appreciation, for some people suggest that policymakers might actually loosen capital controls momentarily and allow Forex to escape in the window around the World Bank disbursement. So again, people looking to move money overseas, taking advantage of this window right around the time of disbursement to do so. Um, and 3rd. Something that Bob referred to as well, that vendors choose to hold potentially illicit and legal revenues offshore for for possibly illicit reasons. I think a key to all of these alternative mechanisms is there's still big news here. We still have a tax havens problem where people are motivated to put money in tax havens, quite possibly for tax evasion reasons, but these stories don't necessarily begin with corruption and diversion of World Bank monies. Hat tip to all the former World Bank economists on Twitter, um. And just kind of continuing on that point, I'm probably. Least solved by what the empirics can give us and what continued regressions can give us to distinguish between these different mechanisms. So, for the corruption and embezzlement mechanism, the paper offers corroborating evidence, you know, that there are bigger effects in more corrupt countries. Yes, but not significantly so, um, that the timing of the effect is, you know, T equals 0, which would be consistent with a politician skimming off of the transfer, but could be consistent with alternative stories. I'd argue the effect is only present for tax havens. I think that's a big important part of the paper, um. But here I want to be particularly careful because I think there's an established assumption in this literature, like in the Zuman, Gabriel Zucman's work, to say that deposits in tax havens are sort of prima facie evidence of tax evasion. That's one step, but that's, it's a big step further to say that deposits in tax havens are evidence that those were ill-gotten gains. That's a different assertion, and I don't think the previous tax havens literature has made that assertion, assertion. So I'm not sure we can use this as evidence that the gains themselves were ill-gotten. Um, and then. You mentioned different people had different priors here. I guess my prior would have been that if this was skimming off of World Bank transfers, I would have expected to see bigger effects for the DPLs than for the investment projects. That's More fungible money, um, and we, we just don't see that here. Um, To wrap up, I mean, I think rather than trying to tease out more of this story with heterogeneous effects, at some points the samples, you know, we're talking about a fairly small number of countries, and I'd be really interested to see some triangulation of the great econometric evidence you have on the core result, with some qualitative support about can you find, Evidence, uh, you know, we have, I was asking former bankers, you know, the Padma Bridge scandal in Bangladesh or the Lesotho Highlands Water project, like, do the known cases of diversion of bank monies fit into the mechanism that you're describing, and can you do some sort of like, you know, you can look for influential observations, we can do the kind of aeronau and Sammy stuff to say, who are the observations that are driving the results. And then I don't know, send it out to the 1818 H Street Club, what is it called, and say, you know, do people have stories about those influential observations? But I, I think a little more qualitative work to back up that mechanism would make the paper a lot, would be fun to read, um, and would make me feel better, um, about this uh corruption mechanism. So in the end, On the sample, if I was going to be super optimistic here, average leakage of World Bank aid is indistinguishable from zero. On the magnitudes, the main takeaway is, hey, aid's a lot better than oil in terms of our leakage rate, and on the proposed mechanism, are we 100% sure that this is really corruption and not something else going on? Let me stop there, but overall, really fun paper, really persuasive main result. Thanks a lot.
showAllTranscripts
no
duration
PT51M32S
scene7File
worldbank/Elite Capture
scene7Domain
https://worldbank.scene7.com/
scene7FileAvs
worldbank/Elite Capture-AVS
title
Elite Capture of Foreign Aid: Evidence from Offshore Bank Accounts
description
Elite Capture of Foreign Aid: Evidence from Offshore Bank Accounts
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
In this seminar on February 28, 2020, World Bank economist Bob Rijkers presented the results of a Policy Research Working Paper on "Elite Capture of Foreign Aid: Evidence from Offshore Bank Accounts."
lp-heading-top-medium
lp-heading-bottom-medium
col-xs-12
col-sm-12
col-md-2
col-lg-2