 Dwi'n meddwl am y cyfnod i'r mod i'r ddisgrifennu ar y Llyfrgell yn yng Nghymru, mae'r ddisgrifennu yng Nghymru yn Yggrifennu Llyfrgell, ond ynghylch yn ffordd, mae'r ddisgrifennu'r ddisgrifennu yn ddisgrifennu, a'r Michael Danker, felly i unig yw yng nghylch yn y Fflaen. Rwyf i'n meddwl, o'r gweithio a chyfnodd, ym mwynhau cyfnod iawn i'r gwbl, sy'n gweld yn enw am syniadau chyfwyr yn cael y tlefiadau. Ond oherwydd ybydd angen fyddiaeth angen i'w ddeallu'n gwybod ni ymddangos gan gyda mynd i'w ddeallu'n gwybod yn fan yw mynd i'r sian o'r cynllun fyddentau'i llyfrgyrchu yn y drywyd ym ni'r treulu'n gwybod a'u chyfyddiad o gref mae cyfnod arnyn yn enw yn cael ei rhôl arall. Ychydigrwyso ar gyfer, rydym nhw rwy'n cael ei gweymau. Felly, rydym ni'n gweld yn oedd gwyun maen nhw, ydych chi'n ffobwyd yn maen nhw i'r llai a ddaeth ei wneud yn gweld yn y ffrif, yw'r pethau, yw mynd i'r wneud, yw'r dweud, yw'r ddod o'r mwyntiau, a mynd i'n ffordd ymlaen i'ch gyrddio'r bwysig, yn cynnig o'r cyfrifiadau yw'r gwaith ar hyn o'r swythio'r cyfrifiadau ar gyfer yna'r ddaeth a'u cyfrifiadau ar gyfer y bwysig ar hyn y bwysig, ac mae'n gweithio'r pwysig o'r cyfrifiadau o'r cyfrifiadau i'r amser o'r cyflwyfridd cydweithio cyfrwyfridd. Felly, rwy'n gweithio am y tro a'n dweud. Mae'n dweud o'r ymddi'r llei Grim Eirben yn ei ddatblygu a'r ddweud o'r cyfrwyfridd yn ein gwaith cyfnodol, mae'r ddweud o'r llwyaf yn i gwyllfa i'r cyfrwyfridd, ac o'r ddweud o'r ddweud o'r ymddi'r llei fyddag fyddai'r llei fyddai. Oherwydd, os ydych chi'n gweithio, Ar yr adran o'r cyfan cyffredineth arlaedwyr celfydio, gyda'r defnyddio'r cyffredineth sy'n roedd cael eu gwirioneddau ar gael cyf hogyol, ac mae'n ddodi i cyffredineth gwrdd cyffredineth. Mae'n ddull gwirioneddol sy'n golygu ar gyfer blaenau cyffredineth, gallw'r cyffredineth, yn rhaid i'ch mynd i ffyrdd y maes yn ôl o'r celfredineth ffyrdd. Mae'r cyffredineth yn ôl o'r cyffredinethau ar gyfer blaenau cyffredineth fel felly rhaid i'ch mynd i ffyrdd o'r cyffredineth ar y cyfnod wgwyrwyr ymwyfyrdd wedi'u adael ar y cyd-wyrdd agynnal wahanol. Yn fawr, bod y bod y cyfnodd difnyddau a'r adael ar y cyfnodd yn yma sy'n byw i gael'r adael fawr, ond mae'r cyfnodd yn fwyaf ar y cyfnodd wedi pwysoch gan cyfnodd yw'r adael ar y cyfnodd. Mae wneud o'r ddefnyddio ar y byddiau ac yn ôl ei wneud. Mae'r ffordd o'r ddefnyddio ar yr arddangos cyfnodd yn cyfrifio'r adael, a rydych chi fod at ychydig i fod yn gwneud y bיעu erfud adrodd byd yn cyrraedd mewn. Rwy'n brom gweithio fel cydweithio i'r fawr i gobl deudio gwybodol i'u gwasanaethau mynd i gynnwys sy'n viadau a midlain yn cael eu gwirionedd. Rydyn ni'n ei chweithio i'r ffordd i gynnwys cyddedigu cyfan hynny, dwi'n bwysig ychydig yw'r cyflwrs yn gweithio fe allwch ar gyfer y cyfrifwyr ac yn mynd i'n fawr any sort of advice that might be in some part based on those scores or figures is grounded in sort of fairly realistic expectations of where countries might see themselves getting to at some point given their underlying characteristics. There are many sort of limitations around tax effort estimation, some of which I'll mention at the very end. O croeiddio, hynny yw'n meddyliaw y cyfrif tegooedd o'r exerciedd a'r ffrind o'r lluniau sgwyd gossip. Fe'r glas y Llywodraeth i ni'n gweld yr exerciedd Pan-draen o'r cyfrif wedi'i ddweud, efallai hefyd yn y cyfnod deallu cael ei ddiweddol o'r ddweud o wladf y lluniau ar y lluniau? Wrth gynllun o'r brif騙wynydd o'r Phaidwyr, oherwydd mae anfermhefnol ar yr OLS, oherwydd mae'n meddyl o'r prynhwниr unedol o'r pryd y lluniau, of a country's tax ratio on a set of economic variables such as the level of income, how open to international trade, the country is the structure of the economy, and maybe some indicator of whether that country has natural resource wealth, which would be perhaps a boost to revenue collections. Over time, studies came to increasingly attempt to understand the sort of mediating role of those factors played by perhaps demographic or socioeconomic characteristics such as how urban is the country, what are the sort of perceptions of corruption in that country, what kind of government is in place, and how that might affect taxes collected across countries. But more recently and where we sort of enter into this debate is that studies have moved to estimating tax effort according to stochastic frontier analysis. I should sort of caveat that of the three of us on the paper, neither Abrams nor I were the geniuses behind the stochastic frontier modelling. Unfortunately, our third author is not in the room with us today, but essentially the stochastic frontier studies started to emerge in the tax effort literature around about 2010, 2013, with some IMF working papers and have been built on a few times since then in the reference studies that are written on the screen there. And this essentially models tax collection as a sort of production function whereby the inputs are those sort of underlying economic and social characteristics, and then the model estimates a sort of tax frontier or a sort of modeled maximum amount of tax that a country could collect given the inputs that go into that model. And essentially in this sort of model, the difference between actual taxes collected in a given country and a sort of theoretical amount or a modeled amount, the difference is broken into a random error term and an inefficiency component. I drew this very basic graph in PowerPoint yesterday, so forgive my artistic skills, but essentially in trying to model this in the most accurate way, we're attempting to hone in as accurately as possible on that inefficiency component and ensure that the sort of random error doesn't creep into what we're actually estimating. So what do we do? Again, I'm giving a fairly rushed or brief overview of what ended up being quite a technical exercise, but we estimate the stochastic tax frontier according to four different models, or we should say our co-author estimates them according to four different models. The pooled model, random effects model, Bateson Co-LE model, and the true random effects approach. And following this estimation, we compute scores of tax effort. And so for as far as we know, our paper was the first to employ the fourth methodology there, the true random effects approach. And we think that that represents a bit of an advancement in our understanding of how to estimate tax effort. So what we did is we estimated, according to these four approaches, the first three have been used at various points in previous papers. And the key question was basically, well, which is actually the best way to model this. And so the next slide is going to show you, after this exercise, the distribution of all of the scores that came out for our model for 160 countries between about 1980 and 2019, the distribution of all those scores according to those four different methods. Hopefully that's not too small or blurry on the screen there, but from left to right, up to bottom, that goes the pooled model, the random effects model, the Bateson Co-LE model, and the true random effects model. And so just focusing on the two at the bottom firstly, the one on the bottom left corner is the methodology that's been used in quite a lot of literature up until now. And you can see that the estimates of tax effort scores range quite broad. The median is around about, I think, 0.35, 0.37. But the true random effects approach seems to stand out from the others as having quite a tight variance and being skewed a lot further to the higher end of the scale with a median at about 0.83 or 0.84. And so essentially we find that in the course of the modelling that the true random effect model is actually better able to disentangle that inefficiency component from the random noise in the model. And the previous approaches that had been employed don't actually seem to be able to do this to as great an extent. And thus some time invariant heterogeneity ends up being attributed to inefficiency, which isn't actually what we want to measure because then you end up with a bigger inefficiency score and it looks like a given country in a given year is not doing as well as it could be at collecting tax revenue. But in fact that there weren't being modelled particularly accurately. And so we find this as a substantive limitation of some of the approaches that had been used before. So in terms of a very, very broad overview of the sort of scores that come out of our model, so we find that globally on average countries are collecting around about 84% of what our model predicts they could be collecting given the underlying characteristics that go into that model. To put that in context with a couple of recent studies, one from 2016 using one of the prior approaches and that I talked about estimates on average a tax effort of about 0.6 for it globally. And I think annually the USAID have a project called the collecting taxes database, data set I think, and they estimate tax effort according to those older approaches as well and find out that an average tax effort globally is around about 0.51, suggesting that an average countries given their underlying characteristics are collecting just about half of the tax revenue that they could be. And sort of just to plot these, so all of the observations we're able to match for each country year of our own approach versus the scores that come from the USAID's collecting taxes database are plotted there. And so you see again on the y axis a much larger spread of scores and really from anywhere from zero to as much as one or in some cases above one actually. So that's just putting in context the sort of differences of what's coming out of the different models. So notwithstanding that we see that the approach that we take comes up with a slightly different set of scores, what we also did very quickly give an example is we looked at what might be causing that bias, what might be causing the fact that in these other approaches some countries are scoring like 99% or 31% or 20%. And when we looked at the input data what we find is that in the previous approaches were being biased by outlying observations of input data. So a very quick example is Slovakia which collects about 20% of GDP in tax revenue, not particularly high or particularly low, but the estimate of tax effort using the Bates Coelay specification is about 0.36. So that suggests that Slovakia is collecting only about 36% of the tax revenue that it could be. We find that actually it's collecting somewhere closer to about 85%. But when we looked at the input variables to try and understand what's causing this we see that trade as a percent of GDP, one of our key economic input variables, has a really, really high figure for Slovakia. It's over 200% for the most recent year and it's one of the highest ranked in the world. We find a similar story for other countries where you had input variables with very high or very low values seemed to be skewing those estimates and these are just sort of like little anecdotal examples but we did see that across countries. So let me just sort of try and tie what all of this kind of means together. Again it was a very sort of quick run through what ended up being quite a technical exercise. We think our results suggest that recent estimates using stochastic frontier analysis of tax effort have perhaps been real substantial under estimates and we think this is due to sensitivity to outlying observations and the methodology that's being employed. And ultimately where these scores enter policy dialogues this can be a bit misleading to go to a country and say you're only collecting 30% of what our model thinks you can collect. It's very different from going to a country and saying well you're collecting about 80-85% of what we think you can collect. Those can be interpreted in two sort of very different ways. But I said I wanted to also just discuss finally about the use of these tax effort scores. It's a very interesting academic exercise and a potentially very useful piece of evidence but I definitely wouldn't ever recommend that these are kind of solely relied upon as a diagnostic tool for suggesting targets or anything like that for revenue collection for a given country. They're very high level. I think they're a useful piece of diagnostic evidence that can play complement to others and that can be used to build a more complete picture of where countries are and where they might hope to go in the future. So other things might be things like tax expenditure analysis which we heard a bit about this morning, losses from initial financial flows, other analyses like VATCAP or income tax analysis can also be very useful tools. And just finally it's worth saying that a tax effort score in a low income country should probably be interpreted very differently to a tax effort score in a high income country. At some point along the sort of chain of growth or development the amount of tax collected, 10 seconds left, the amount of tax collected essentially becomes a societal choice of what kind of party you voted for and whether there are a low tax or a high tax party. But I think it's probably fair to say that most low and middle income countries would just like to collect a lot more tax revenues to fund the development spending. So interpreting a score of 0.85 for Malawi or Sierra Leone might be a very different interpretation to a score of 0.85 for Norway or for Sweden. Thank you and apologies for going over time.