When you compare apples with pears and then bananas you might get a half decent fruit salad but you will definitely get terrible analysis.

In a recent conversation, Jonathan Portes, Professor of Economics and Public Policy at King’s College London, informed me that the Office for Budget Responsibility (OBR) forecast for the cost of Brexit was “always only trade, and they’ve always modelled migration separately”.

The trouble is I know that is both factually incorrect and also only the tip of the iceberg of the issues with the OBR Brexit analysis.

Whilst it is true that the OBR state that their forecast for lower GDP outside the EU is based on a 4 per cent decline in productivity underpinned by a decline in trade intensity, to quote their Brexit analysis (updated 17th April 2023):

The post-Brexit trading relationship between the UK and EU, as set out in the ‘Trade and Cooperation Agreement’ (TCA) that came into effect on 1 January 2021, will reduce long-run productivity by 4 per cent relative to remaining in the EU. This largely reflects our view that the increase in non-tariff barriers on UK-EU trade acts as an additional impediment to the exploitation of comparative advantage.

The OBR have not reached this figure via their own in house economic model. As they explain:

It is not within the OBR’s mandate or resources to discern and separately estimate the impact of the UK’s exit from the EU and related policies on all the elements of our forecasts on an ongoing basis.

Rather, they have reached their conclusion instead by looking at 13 modelled outcomes from 11 external providers and then calculating an average. As they state:

 In order to generate this figure, we looked at a range of external estimates of the effect of leaving the EU under the terms of a ‘typical’ free trade agreement (FTA) (see Box 2.1 of our March 2020 EFO for more information).

This is the table of external providers’ forecasts the OBR refers to, with the 4 per cent average figure at the bottom. 

The trouble is, whilst the OBR might claim that its analysis is based only on a decline in productivity underpinned by a decline in trade intensity, several of the studies they have included in their average do include a negative Brexit effect from immigration in their models.

For example, included in the OBR table is the NIESR model which predicts the UK economy to be 3.8 per cent smaller by 2030 as a result of the UK leaving the EU (with a free trade agreement). However, having spoken with the NIESR about the details of their forecast, they make clear that this 3.8 per cent includes forecasts for lower immigration. On a per capita basis their estimate is only 3.0 per cent lower.

Similarly Felbermayr et al in the OBR table (Felbermayr, Fuest, Groeschl and Stöhlker), in their piece “Economic Effects of Brexit on the European Economy”, forecast 1.8 per cent lower UK GDP versus staying in the EU by 2030. However, again this includes a forecast for lower immigration. In fact the majority of this lower growth forecast is due to an estimate for lower immigration. On a per capita basis, its forecast is for only 0.6 per cent lower GDP in the event of an FTA deal.

In short, the OBR average of 4 per cent is arrived at by comparing apples with pears.

You might also notice in this table that there is a wide range of forecast outcomes. From Felbermayr et al — a per capita fall of 0.6 per cent — to the World Bank forecast of 10 per cent. Indeed of the 13 external forecasts referred to in this model, 9 (70 per cent) have produced an estimate which is below the 4 per cent average which the OBR uses as its cost of Brexit. Indeed if you exclude just one of the estimates (the World Bank — which is 2.5 times higher than the average), the arithmetic average falls to about 3.5 per cent.

Anyone who knows anything about standard deviation would understand why such outliers are typically excluded from such analysis (although there are other very good reasons to exclude the World Bank estimate from the figures as I shall explain later).

Another thing you may notice when studying the table is that the outcomes are very different depending on the model type used. Those which use “constant returns to scale” tend to have lower Brexit costs than those who use “dynamic productivity”.

This is because the “constant returns to scale” models are (broadly) attempting to assess only the costs of the non-tariff trade barriers erected by the UK leaving the EU, whilst the “dynamic productivity” models are also trying to calculate a potential loss of productivity to UK PLC as a whole by leaving the SM & CU. The argument being that the “cost” is not just the time and expense and opportunity cost of having to complete new customs paperwork, but also an overall reduction in productivity to the WHOLE economy from less integration. This is what the OBR means when they say that their 4% Brexit hit is driven by a decline in productivity underpinned by a decline in trade intensity.

Take the NIESR figure of a 3.8% hit. As we have already discussed, 0.8% of this is derived from expectations of lower immigration. But they also informed me that the single biggest factor was their forecast for lower productivity — the impact of lower productivity alone represents 2.1 per cent of the 3.8 per cent forecast decline. This means that just two variables — lower productivity and lower immigration — are responsible for 2.9 per cent of the lower growth forecast. Or to look at it another way, the forecast for lower GDP driven only by the loss of single market and customs union membership / increased trade friction, is just 0.9 per cent.

Or take the forecast of the Netherlands CPB in the OBR table as Computable general equilibrium (Krugman style increasing returns to scale). They forecast a Brexit hit of 3.4 per cent including “trade-induced innovation losses” in the event of an FTA.  As their table below shows, this is higher than the 1 per cent estimate of the LSE and 1.2 per cent estimate of the PWC because, “The PWC estimation of costs under the FTA scenario…does not include any additional losses from less innovation due to less trade.”

So, the Netherlands CPB’s 3.4 per cent Brexit hit already includes at least 2 per cent from “a decline in productivity underpinned by a decline in trade intensity”. That must mean that their second entry in the OBR table, under the title computable general equilibrium (dynamic productivity), which forecasts a 5.9 per cent hit, includes an almost 5 per cent hit from productivity alone. 

UK in a Changing Europe is also in the OBR table twice — once with their “constant return to scale model” which is derived from the Centre for Economic Performance (CEP) trade model and once with their “dynamic productivity” model. They forecast a Brexit hit of 2.5 per cent.

However, as they go on to explain “The CEP trade model does not allow for any dynamic effects of trade on productivity”. Including that their Brexit loss rises to 6.4 per cent.

And one last example. In the OBR table Mayer et al 2018, the Brexit hit is recorded as 2.4 per cent. However, as the table below taken from their report illustrates, trade friction alone is forecast to “cost” 0.8 per cent rising to 2.4 per cent including “reduced trade openness” effects.

So what have we learned so far? First, the OBR forecast of a 4 per cent Brexit hit to the economy is not derived from its own economic model but an average of 13 other models from 11 different external providers.

That whilst the OBR states that its forecasts exclude immigration, some of the models included in its average do include immigration.

That whilst the OBR states that its forecast is based upon “a decline in productivity underpinned by a decline in trade intensity”, not all the models included in its average calculate their Brexit effect in this manner and those who do have a very wide range of modelled outcomes.

That from these external providers, the forecast for lower GDP driven only by the loss of single market and customs union membership / increased trade friction alone is considerably lower than the OBR 4 per cent estimate. Indeed, using the analysis above it is possible to present the following data for the average forecast for lower GDP driven just by increased trade friction. That is to say, excluding a forecast negative productivity or immigration effect.

Modeller	Increased trade friction cost alone (%GDP)
Felbermayr et al (2018)	-0.6
Mayer et al (2018)	-0.8
NIESR (2018)	-0.9
LSE	-1.0
PWC	-1.2
CPB Netherlands	-1.2
IMF (2018)	-2.0
UK in a Changing Europe (2019)	-2.5
Average	-1.3

As you can see, the majority of the forecasts for lower GDP driven only by the loss of single market and customs union membership/increased trade friction are around 1 per cent with the UK in a Changing Europe a significant outlier at 2.5 per cent. 

Should a decline in productivity underpinned by a decline in trade intensity be included in forecasts?

As I explained above, the OBR’s (and others’) argument is that the “cost” of Brexit is not just the time and expense and opportunity cost of having to complete new customs paperwork, but also an overall reduction in productivity to the whole economy from less integration. 

There is some logic to this argument. Intuitively, one would think that companies which are more open to alternative ways of operating and have access to a wider pool of suppliers, customers and capital providers should be more innovative. Yet contrary to the arguments of some, the academic literature and analysis on this subject is by no means definitive.  

As Bhattacharya, Okafor and Pradeep state in their research, entitled “International firm activities, R&D, and productivity: Evidence from Indian manufacturing firms”:

Although the links between productivity and international firm activities, including exports, trade in intermediate inputs and capital goods, and foreign direct investment (FDI), have been studied extensively, the evidence on these relationships is mixed.

Or as Patricia Hofmann explains it in her research looking at Germany “Export-Intensity and Productivity Growth: Evidence from German Firm-Level Data”:

While with these improvements the actual channels and mechanisms of how trade (liberalisation) may influence economic growth and technological change are better understood, the main conclusion that can be drawn is that there are long-run growth and welfare effects of trade, but the sign of these effects is ambiguous.

Rodriguez and Rodrik in their paper for the NBER entitled “Trade Policy and Economic Growth: A Skeptic’s Guide to Cross-National Evidence” go further: 

We find little evidence that open trade policies — in the sense of lower tariff and non-tariff barriers to trade are significantly associated with economic growth …

In the UK, the Office for National Statistics (ONS) in 2018 released the following work ,“UK trade in goods and productivity: new findings”, which intended to shed some more light on the subject for the UK. Their analysis is based on a new dataset which uses HM Revenue and Customs’ administrative trade data linked to business-level financial data to analyse the link between productivity and trader status for British businesses.

They found that businesses that trade (import/export) outside the UK do have a higher productivity, although the “productivity premia” differed by geographical market:

These results suggest that businesses which export to (import from) the EU are around 4.3% (1.7%) more productive than otherwise equivalent non-traders. However, these effects are much smaller than the equivalent estimates for non-EU trade: businesses which export to (import from) non-EU nations are around 19% (18%) more productive than equivalent non-traders

Importantly, they add:

In particular, our results are not causal: they cannot say whether these businesses are more productive because they trade or whether they trade because they are more productive.

So, the academic evidence regarding trade openness and productivity is pretty mixed. Even those who do determine a higher productivity for companies who export/import, many seem reluctant to determine where this is a result of their trading abroad or the reason why they trade overseas in the first place. There also appears to be a far greater productivity premium for those who trade outside the EU.

But you may also have recognised a further issue with the academic analysis. It predominantly covers trade in goods and not services. That is not really a surprise. As the Bank of England states: 

Trade liberalisation has been asymmetric, focused on goods rather than services trade. The decline in goods trade barriers may have favoured countries specialising in goods, like China, Germany and Japan, allowing them to increase exports relative to imports, and contributing to their persistent current account surpluses. By contrast, countries like the United States and the United Kingdom, who specialise in the services sector where trade is more restricted, have been running persistent deficits.

For some countries this may not matter too much — goods exports tend to significantly outweigh services — but the UK is a global outlier with over half of all exports being derived from services versus an average of only 20 per cent for the world (7 per cent for China) and just over a quarter for the G7 (source OECD 2020).

Meanwhile, even those who do find a causal link between trade openness and productivity are widely divergent in their estimates of just how important it is. The Department for Business and Trade’s research paper “The relationship between trade and productivity: a feasibility study”, which relies heavily on the ONS data above, is convinced that “engaging in importing and exporting is associated with a subsequent increase in TFP (total factor productivity) of 6.7 per cent amongst UK firms”. However, they also state that:

At the UK – national and sectoral level, there is a lack of evidence on what role trade can play in facilitating productivity gains … At the firm-level, data issues have rendered simple statements like “if trade goes up or down by x% then UK productivity increases/decreases by y%” difficult to infer.

And that brings us to the next problem with the external economic models the OBR use to reach their 4 per cent forecast Brexit hit. Because these models do make calculations based on simple statements like “if trade goes up or down by x% then UK productivity increases/decreases by y%”.

Take the UK in a Changing Europe for example. They state that for their dynamic productivity model they “turn to the empirical literature on how trade affects income per capita. A central estimate from this literature is that a 1% decline in trade reduces income per capita by around 0.5%.”

In other words, the difference between their forecast of a 2.5 per cent Brexit hit through trade friction alone and their 6.4 per cent forecast through productivity losses is literally as simple as “if trade goes up or down by x% then UK productivity increases/decreases by y%”.

With manufacturing representing only 10 per cent of UK GDP (and less than half of exports) versus more than double the GDP contribution for Germany (and over three quarters of exports), you can quite understand how ludicrous the idea is that there could exist one magic formula or global equation to measure the productivity effect on all countries from trade openness. 

Considering the lack of consensus in the empirical literature outlined above, considering the scale of the difference this makes in the forecasts and considering how the pronouncements of the OBR control/limit Government policy, this seems an extraordinary assumption to make.

What does the actual trade data tell us?

Other external forecasters that the OBR rely on for their 4 per cent Brexit cost have analysed the trade and productivity hit in a more granular manner. This enables us to look in detail at some of their assumptions and compare them to the real world actual performance. For example, the World Bank, which you will remember forecast a 10 per cent reduction in UK GDP from leaving the EU with an FTA.

They state (see table below): 

Estimates suggest that the sharp drop in the agreement’s depth between the UK and the rest of the EU would lead to a 38 percent drop in (UK to EU) gross exports of goods…The largest decrease would be for exports of services that are estimated to fall by 48 percent.

How does this compare to what has happened in the real world?

Well according to the ONS at the end of 2023, UK goods trade was around 10 per cent below 2019 levels whilst UK services trade was around 12 per cent above 2019 levels (by far the strongest growth in the G7). In terms of overall trade intensity (exports plus imports) in the third quarter of 2023, UK trade intensity remained 1.7 per cent below its pre-pandemic level from 2019. 

As I am sure you can see. In terms of scale, the World Bank numbers aren’t just a little bit off. They aren’t just in a different ball park. They are on a different planet! 

Which brings us to our next issue with the OBR Brexit forecast.

The OBR state that: “Since the June 2016 EU referendum, our forecasts have assumed that the volume of UK imports and exports will both be 15 per cent lower than if we had remained in the EU.”

There is certainly evidence that trade intensity in goods has fallen (although how much is due to Brexit is a moot point). But when it comes to services it is quite clear that the UK is outperforming considerably. As we discussed above with more than half of exports coming from services this matters a lot. Indeed the UK performance in services is even more impressive when you consider that for the whole of the OECD exports of services declined by more than twice the value of goods exports as a result of Covid lockdowns and have since been slower to recover.

It is clear that real life actual trade data does not support the OBR forecast of a 4 per cent Brexit hit, but more than that their “modelling” does not take into account any potential benefits from leaving the EU (which probably should tell us something).

For example, trade intensity measures both exports and imports. And the data shows when it comes to goods, both are lower. But this isn’t a result of lower UK consumption. Rather what we are witnessing is UK manufacturers capturing a larger share of domestic consumption.

For example, since Q2 2016 (Brexit referendum) UK manufacturing volumes have grown considerably faster than all peers.

And in fact as a result, the UK is the only one of its peers where manufacturing volumes are higher now than in 2008 before the Great Financial Crash.

Even if one was to argue that this growth in UK manufacturing is less productive than the imports it replaces — and considering growth in UK manufacturing as a result of Brexit  was not included in any of the economic models on which the OBR’s forecast is based — surely the OBR should make some (positive) adjustment to its 4 per cent estimate? After all, when it comes to immigration, the OBR is able to add to GDP growth even if their forecast for GDP per capita (productivity) as a result is lower. It rather seems that when it comes to Brexit, the OBR only considers one side of the Profit and Loss statement.

Does reshoring reduce productivity?

But it is far from clear that this reshoring is actually less productive than the imports they replace. The academic research we studied earlier is all from a time when manufacturing was being offshored to cheaper locations. When globalisation was in its ascendancy. That is not the case now. Now it is reshoring — defined as the relocation of previously offshored production activities back to the home country — which is in vogue. Not least because of the supply chain disruptions caused by Covid. But academic studies are few and far between, so recent is this phenomenon.

One such 2023 analysis by Luca Pennacchio entitled “Reshoring and firm productivity” investigated the impact of reshoring on the productivity of manufacturing firms using data from the European Reshoring Monitor on European firms in the period 2014–18. 

His main findings were that reshoring increased productivity in small and medium-sized enterprises (SMEs), especially when they repatriated production from distant and developing countries but that there were no significant productivity changes (up or down) in large firms or firms repatriating production from more developed countries in America and Europe.

Meanwhile, Suraksha Gupta, Yichuan Wang, Michael Czinkota in their 2023 work “Reshoring: A Road to Industry 4.0 Transformation” found that by minimising reliance on human labour, automation results in consistent, high-quality outputs and enhanced productivity when reshoring.

A Deloitte global study found that companies running smart factory initiatives on average saw a 10 per cent increase in production output, 11 per cent increase in capacity utilisation, and 12% increase in labour productivity and that Smart factory capabilities provided opportunities for value creation at both greenfield and brownfield facilities.

According to Manufacturing Logistics and IT, 90 per cent of manufacturers in the UK engaged in the reshoring process report successful outcomes. Around a quarter report enhanced value, heightened security and lower costs. According to their polling research “business leaders say AI driven tracking platforms (31 per cent), digitalisation in factories (29 per cent), and data analytics for risk management (29 per cent) — i.e. smart factory capabilities — are enabling the move towards reshoring.”

There are plenty of other studies which argue that AI and robotics (industry 4.0) could transform UK productivity and GDP. PWC argue UK GDP could be boosted by over 10 per cent (the equivalent of an additional £232bn). The UK Government’s Made Smarter report argues that the rise could be as large as £455bn (18 per cent).

A September 2022 survey by the Chartered Institute of Procurement and Supply reported 45 per cent of companies were planning to move their production back to the UK in response to challenges faced during importing and a consistent increase in shipping costs. An October 2023 survey by Censuswide for Manufacturing Logistics and IT found 58 per cent of manufacturers have started to do so. 

It is clear that reshoring is happening, and in a very significant way. But what is not yet clear is what effect this will have on productivity. Certainly based on the available evidence it would seem incongruous to assume a productivity decline. 

So what does current productivity data tell us?

We have already looked at current trade data which shows that the OBR’s thesis that trade intensity will be 15 per cent lower than as EU members is nowhere near being demonstrated in the data. But what does current productivity data tell us about what is happening? Remember the OBR’s thesis is that a drop in trade intensity will result in a 4 per cent drop in GDP per capita (productivity).

If the OBR is correct we should see the beginning of a divergence between the UK and Eurozone when it comes to productivity. After all, the OBR argument is that Brexit will cause a hit to UK productivity, not the EU. However as this Chart from the FT demonstrates the opposite is occurring. Since 2016 and the Brexit referendum and from January 2021 and the end of the transition period  the gap between UK and EU productivity has closed as UK productivity has continued its slow rise and Eurozone productivity has flat-lined/drifted lower.

The real story on productivity (as I explained here and the FT has since belatedly realised) is the whole of Europe’s “competitiveness crisis” versus the USA and how this productivity gap has been widening since 2008. Not since 2016 and the Brexit referendum, and not since 2021 and the end of the transition period. It is obvious that the UK (and the rest of Europe) has a productivity problem; but the data does not suggest that problem is as a result of Brexit.

In fact I agree with the OBR when they state that:

 … the economic and fiscal shock associated with the end of the Brexit transition period occurred as the country was still dealing with the pandemic. This comes on top of the stagnation in productivity seen since the global financial crisis. And more recently, the impact of the Russian invasion of Ukraine. These factors mean that it will, in practice, be very difficult to isolate the effect of the new trading relationship with the EU on the medium-term economic and fiscal outlook.

Yes indeed. Difficult to isolate. Difficult to prove. And as I hope I have demonstrated, difficult to justify too.

In March 2018, an important commentator on this Brexit debate made the following statement (apologies for publishing it in full but I think you will understand why later): 

There is a degree of consensus that leaving the EU will result in greater trade frictions in aggregate and that increasing trade frictions will reduce openness. But there is much less agreement on whether, and by how much, reducing openness will affect productivity directly – for example, this channel was an important factor in the Treasury’s pre-referendum analysis, but NIESR chose not to include it. The Dutch fiscal council argued in its pre-referendum analysis that “Quantifying these dynamic effects has proven difficult, for two reasons. In the first place, it is difficult to capture the link between trade, knowledge transfer and innovation as one specific mechanism; the relationship is much more complex. Therefore, it is not easy to include in trade models. In the second place, empirical studies quantifying the effect are proven to be faced with a number of econometric problems.

The empirical evidence regarding the impact of openness on productivity is mostly drawn from cross-country growth regressions, where much of the information in the data derives from increasing trade intensity in developing countries. That experience may not be relevant to an advanced economy like the UK. There are also econometric qualifications attached to many of these studies, including the possibility that the openness measures may reflect the influence of omitted factors that drive cross-country productivity growth differences. Finally, there are issues as to how openness is measured and whether the estimated elasticities can be applied to countries with a very different composition of trade; for example, the UK’s share of services in total exports is higher than in most countries and global trade has been liberalised less in services than goods.

Moreover, much of the evidence relates to increases in openness and rather less to reductions, as would be the case with Brexit, and there may be asymmetries in the impact of changes in trade frictions. For example, one of the ways in which increased openness is thought to increase productivity is through knowledge spill overs, but reducing openness by introducing trade frictions should not lead businesses to forget what they already know. Finally, it is plausible that the productivity consequences of changes in openness will only become manifest over quite a long time horizon, certainly beyond our current five-year forecast limit.

For these reasons, we have chosen not to incorporate an explicit link from lower trade intensity after Brexit to lower productivity growth within our forecast horizon.

Sounds like something I could have written. But it wasn’t me. Who was the author? 

Why, it was the OBR…. 

That’s right. In March 2018 the OBR itself stated that it would be wrong to “incorporate an explicit link from lower trade intensity after Brexit to lower productivity growth within our forecast horizon”.

The OBR were right to say that back in March 2018. They would be right to say that today too.

It is clear from the analysis above that there is an economic cost to leaving the EU single market and customs union. It is also clear that this direct cost is in the region of 1 per cent of GDP. I agree with the OBR, assuming any other effects — incorporating an explicit link from lower trade intensity after Brexit to lower productivity growth — is simply not reasonable.

The OBR’s Brexit analysis is fundamentally flawed. Not only does it compare apples with pears, not only does it exclude any positive economic effects from leaving, it also includes unreasonable and unjustified predictions for productivity. 

The OBR should take its own advice and admit that the only costs of Brexit which can be safely and reasonably modelled and assumed are the direct costs. And that these direct costs are in part mitigated by post Brexit reshoring of production. The rest isn’t just apples and pears but unproven bananas.