Distributional house price cycles (2/2): What explains the divergence?

Posted on October 10, 2026 by Francesc Rodríguez Tous

In the previous post, I discussed the building of the distributional house price indices (DHPIs) for the U.K., noting that they allow us to study something that is typically not possible with other house price indices: the within-local market heterogeneity in house price growth. In this second (and last) post, I will go through the main findings of the paper: the heterogeneity, in particular the differences between low- and high-priced houses, is systematic.

Before going there, however, let me mention something that is crucial for the analysis: all the results are within-market statements. In econometric terms, I use local area-by-quarter fixed effects. That means that shocks affecting areas are captured by these fixed effects. One question that comes from this approach is whether there is meaningful variation within markets. The answer is yes. A simple regression of house prices with local area-by-quarter fixed effects shows that most of the variation in house prices remain unexplained. So my approach controls for local-area trends while focusing on substantial variability.

The first question that I had from plotting the DHPIs was whether different parts of the house price distribution move differently depending on the position of the economic cycle. Using (lagged) real GDP growth, it turns out that they do: when GDP growth is high, lower-priced houses increase by more than higher-priced ones. The magnitude is important: a one standard deviation increase in GDP growth is associated with 0.46 percentage points higher price growth for cheaper houses compared to more expensive ones. What is also important is the monotonicity of this relation: as we move closer to the right of the price distribution, this heterogeneity dies down.

GDP growth is a measure of the economic cycle, but another one that might be more relevant is credit. Indeed, macroprudential regulators tend to look at measures of credit to understand whether they need to use macroprudential tools. So I repeat the same exercise using net secured lending flows, that is, how much new secured lending (basically mortgages) banks and other financial institutions provide minus the repayments during that quarter. The same relation appears, but with a twist: cheaper houses increase in prices by more when credit is high, but the magnitude is almost three times stronger than for GDP growth (1.24 vs. 0.46). Indeed, when we put both measures of the cycle together, the coefficient for GDP growth drops substantially and is no longer significant, while the coefficient for credit barely changes. In other words, the cycle that matters for the heterogeneity is the credit cycle.

That is all great (well, at least that’s what I think) but it is about correlations. Is the cycle causing the heterogeneity? You could construct an alternative story. For instance, when households in the lower-priced housing segment expect this segment to appreciate substantially, they borrow more, which pushes credit up, and subsequently are more likely to buy houses, pushing demand and prices up. It would be great to use exogeneous shocks instead. This is where monetary policy surprises come in.

Monetary policy decisions are extremely endogenous to the conditions of the economy. But these decisions are also incredibly consequential for businesses, financial institutions, households, and macroeconomic performance. So the literature has for many years now spent a lot of time trying to obtain monetary policy shocks. One of the most used approaches is using monetary policy surprises, that is, changes in yields around monetary policy announcements. This is what I use here, partly due to the availability of these shocks for the U.K. from Braun et al. (2023).

What should we expect if the channel is that higher credit availability leads to higher housing demand particularly for constrained borrowers, who are more likely to be in the cheaper segments? We should expect that higher interest rates are associated with lower price growth for cheaper houses. This is indeed what I find. Positive monetary policy surprises (that is, monetary policy announcements that revise yields up) reduce the price growth of cheaper houses compared to more expensive ones. The effect comes from changes in the short-end of the yield curve, rather than changes in longer-term yields, consistent with the channel coming through mortgage lending. I also show that the effect does not revert: the drop in price growth continues to accumulate for at least four years after the monetary policy shock.

There are other results and details that I cannot include in the blog, so I encourage anyone interested in the topic to look at the paper. Also, for the first time in my academic career, I have included a replication package on GitHub; perks of not using confidential data. This is just version 1; I am extending the analysis in two directions now: studying quantities and bringing local market heterogeneity. But so far, I think the results bring something that may not be well-known: within-market heterogeneity is substantial, and different price segments move systematically with the credit cycle. Importantly, monetary policy has distributional consequences within local markets through house prices, hitting lower-priced houses the hardest.

Posted in Uncategorized | Tagged , , , | Leave a comment

Distributional house price cycles (1/2): Do cheap and expensive homes move together?

Posted on September 28, 2026 by Francesc Rodríguez Tous

Within a local housing market, do the cheapest and the most expensive houses rise and fall together over the cycle? One would have expected this to be a well-documented fact, but it turns out to be surprisingly difficult to answer with the data we usually have.

House price indices typically present one number per region or local market. I can find out how much house prices grew in Chelsea last year, but not whether the bottom of the Chelsea market grew more or less than the top. As a result, most of what we know about heterogeneity in house price dynamics is about differences across places (London against the North, cities against towns), and much less about differences within them. Why is that not enough? Differences across regions can be the result of many things, such as capacity for building new housing, regulations, local economic activity, etc. It is difficult, therefore, to pinpoint the heterogeneity to a particular factor when comparing different regions.

A naïve approach (and I do not mean this in a derogatory way; that’s exactly what I did some years ago in a project) would be to simply compute percentiles of the house price distribution within each area and study how this distribution evolves. What’s wrong with this? The problem is that the houses that go to the market change with the cycle. For instance, imagine that cheaper houses tend to go on the market more often during booms. If I am simply tracking the distribution across time, it might look like it moves to the right during a bust. The key distinction is that we are interested in the price of housing. We need to isolate changes in the price of housing from changes in the composition of houses that go on the market.

The standard ways of dealing with this are to follow the same house across sales (repeat-sales indices) or to adjust prices for the characteristics of the property (hedonic indices). Neither is easy to apply here. The same house rarely sells twice within a short window, so repeat-sales samples become thin in a local market, let alone, as we want to do here, in one segment of it; and hedonic indices require characteristics that transaction data generally do not contain.

In a recent paper, I try a different route, which relies on a peculiarity of the UK: its full postcodes are very granular. There are almost two million of them for fewer than thirty million households, so an average postcode covers around fifteen homes. I define a unit as a postcode-house type pair (say, terraced houses in a given postcode) and, each quarter, I keep the units that also transacted a year earlier. That seems close enough to “the same house” for this purpose. Within a local authority, I then compute the percentiles of prices over exactly that set of units in both quarters and take the growth rate. Since the basket is the same at both ends, the change should reflect the price of housing at that point of the distribution rather than a change in which houses happened to sell. I call the result distributional house price indices (DHPIs), and I build them for the 20th, 40th, 60th and 80th percentiles of every local authority in England and Wales from 1996 to 2019 (the paper also uses finer sets of percentiles; the results are very similar). Figure 1 shows the average across local markets of the four indices.

Figure 1: DHPIs P20, P40, P60, and P80 year-on-year growth, cross-market average

The four lines move together, which is reassuring: the indices seem to pick up the cycle we all know. But they do not move by the same amount. At the top of the boom, in early 2003, the 20th percentile was growing at around 30% a year against 26% for the 80th. At the bottom of the bust, in early 2009, it was falling at 16% against 13%. In calmer years, the lines almost coincide. Figure 2 plots the difference between the growth of the 20th and the 80th percentiles, which makes the pattern easier to see.

Figure 2: Spread between P20 and P80 year-on-year growth, cross-market average

The spread is clearly pro-cyclical. Through the 2002–2005 boom, it averaged close to five percentage points a year, peaking at nine points in 2004. In the 2008–2010 bust, it turned negative, reaching almost minus four points in early 2009. Over the whole sample, however, it averages less than half a point and is positive in only slightly more than half of the quarters. Therefore, this does not look like a story about cheaper houses being a better (or worse) investment; it looks like a story about cheaper houses having a more volatile cycle.

All of this is within a local market. Still, one could worry that the averages are driven by a handful of places, or by the familiar divergence between London and everywhere else. Figure 3 maps the two halves of the global financial crisis separately: for each local authority, how much the bottom of the distribution outgrew the top during the 2002–2006 expansion, and during the 2007–2011 contraction.

Figure 3: Cumulative P20–P80 gap by local authority, GFC, boom (left) and bust (right)

In the expansion, cheaper houses outgrew more expensive ones in 93% of local markets. In the contraction, they fell behind in 85% of them. More than three quarters of local authorities are red on the left map and blue on the right. The cyclical differences within markets, then, do not appear to be simply a regional phenomenon: something similar happened in most markets of the country.

What about the long run? Between the mid-1990s and the late 2010s, prices roughly quadrupled in the typical local market, while the cumulative gap between the 20th and the 80th percentiles amounts to only a couple of log points on average. The boom and the bust seem to have largely cancelled out. That average, however, hides a fairly clear geography, as Figure 4 shows. The bottom of the distribution gained the most along the Thames corridor and the outer commuter belt and in outer and east London. It lost the most in the northern cities, and in general, the North of England saw the top outperforming the bottom of the distribution.

Figure 4: Long-run P20–P80 gap by local authority, 1995 – 2019

The paper focuses on the cyclicality of the gap (Figure 2) rather than the long-run differences. In the next post, I will go over the determinants of this cyclicality. But seeing the shape of the gap, a good candidate easily comes to mind: credit.

Posted in Uncategorized | Tagged , , | 1 Comment

Borrower-based macroprudential tools (4/4): The need for transparency

Posted on July 29, 2026 by Francesc Rodríguez Tous

In this series of posts, I have made the case for borrower-based macroprudential policies, discussed their distributional effects, and compared them with monetary policy. One overarching theme has been that the costs of these tools are immediate and distributional, while the benefits, although most likely distributional as well, are much more difficult to quantify.

Quantifying these benefits requires serious empirical work, detailed household- or loan-level data, and, often, a negative shock such as a recession. The key issue is that these data are typically held by central banks. The arrangements differ across institutions, but access by outsiders is usually highly restricted. The same institution that designs and implements these policies is therefore, in many cases, also the one best placed to evaluate them.

It is not difficult to see why that can be a problem. There is evidence that research done within central banks tends to report stronger effects of quantitative easing (compared to non-central-bank researchers). There have been some papers disputing this claim, and the authors have responded. There have also been more direct allegations of institutional interference elsewhere: The World Bank allegedly tried to suppress a paper documenting elite capture of foreign aid.

And these are only the instances we know about. We cannot know how often preliminary findings have been discouraged, delayed, or abandoned because they went against an institution’s official position. Nor does an institution need to stop an existing piece of research. Sometimes central banks announce their research priorities and, once a policy is in place, its evaluation simply never appears among them. There can be a lot of “soft management” to steer research away from certain topics.

So we have, on the one hand, an institution with considerable power that uses policies whose costs are unequal and whose benefits are difficult to quantify. On the other hand, the data necessary for this evaluation are controlled by the same institution and remain largely inaccessible to outside academics. The conclusion is, unfortunately, clear: we should not have to rely exclusively on in-house analysis. Not because the analysis is necessarily wrong, but because institutional incentives may affect which questions are asked and how the evidence is interpreted, while outsiders often have no way to replicate, challenge, or extend it.

So what should central banks do? For one, they should open up their data. Obviously, they cannot make all of it public. But they should provide secure ways for external researchers to use the data and analyse these policies. This should not require co-authors from the central bank, nor should the central bank exercise substantive oversight over the research question, the methods, or the conclusions. Its oversight should be limited to protecting confidentiality, ensuring data security, and preventing identifiable information from being made public. Some central banks already provide this type of setup, for instance the Deutsche Bundesbank, but these arrangements remain uneven and often apply to a small proportion of the data that they collect. The empirical work of assessing the policies—and not just borrower-based macroprudential tools—could largely be carried out by researchers themselves, without central banks having to commission (and pay for) it.

This would have other desirable consequences. To date, if one wanted to conduct research using loan-level data, one often needed a contact within the central bank (the publicly available US mortgage data collected under the Home Mortgage Disclosure Act being a notable exception). I have several papers using confidential central-bank data, and this has always been my experience. This is understandable, given the confidentiality of the data, but it also means that researchers without those contacts are at a disadvantage when trying to do research in banking and financial markets. Providing access to external researchers in the way suggested above would help level the playing field.

But the key reason would be to promote transparency: using the academic community to learn more about how these policies work and to provide independent evidence on their costs and benefits. Central banks may be best placed to collect the relevant data, but they should not be the only institutions able to analyse them.

This is, of course, far from the only problem central banks are facing at the moment. Political pressures to curtail their independence and the questions about their legitimacy are genuine issues that need to be tackled. But somehow I do not think that becoming more closed will help. Nor do I think that the way forward is for central banks to assess the impact of their actions less carefully, or to make it harder for others to do so. If central banks want to preserve the powers and independence they have been granted, they need to make a stronger case for them. Making the relevant data available, and allowing outsiders to examine whether their policies are working, is part of that case.

Posted in Uncategorized | Tagged , , | Leave a comment

Borrower-based macroprudential tools (3/4): When success is hard to observe

Posted on July 27, 2026 by Francesc Rodríguez Tous

In the first two posts of this series (here and here) I have argued that the case for borrower-based macroprudential tools is sound and that, while they appear to be effective, they have distributional costs. I finished the previous post mentioning that the distributional benefits are much more difficult to estimate than the costs, although there are good reasons to think that they are there. In this post, I want to consider what this asymmetry means for how these policies are conducted. One way to do so is to compare them with one of the main roles of central banks: monetary policy.

I like this comparison, because monetary and macroprudential policies are more similar than one might expect. For instance, we have been discussing how macroprudential policy can prevent some households from buying their first house. But the same is true with monetary policy: higher interest rates increase mortgage rates, which can prevent lower-income households from accessing this type of lending.

One difference worth noting is that monetary policy affects both potential and current borrowers, at least in countries with a substantial proportion of adjustable-rate mortgages (or with short fixation periods like the U.K.). Borrower-based macroprudential policies fall much more directly on potential borrowers or, more precisely, on people who want to buy a house via borrowing.

Monetary policy is therefore also distributional. We have a substantial body of work on this question, both theoretical and empirical (Auclert, 2019; Cloyne et al., 2020). Changes in interest rates redistribute income and wealth between households depending on their indebtedness, asset holdings, and sources of income. In particular, borrowers with adjustable-rate debt tend to lose when interest rates go up, while net savers tend to gain. Through this direct cash-flow channel, the effect can reverse when interest rates fall. This symmetry is much less evident when macroprudential limits are introduced. A household prevented from borrowing to buy a house today is not automatically compensated when the limit is relaxed several years later; by then the household might not be financially constrained anymore.

But there is another key difference between the two policies. It is subtle, but I would argue that it has significant consequences: the feedback available to policymakers and the public.

Monetary policy targets inflation. Inflation figures are published frequently, and the target is usually explicit. We can therefore observe whether inflation is above or below the target and whether it appears to be moving in the right direction. Note that this does not allow us to determine scientifically (i.e., to “prove”) whether the central bank is succeeding. For that, we would need to observe a counterfactual world (or estimate one) in which the central bank had followed a different monetary policy and compare the resulting inflation. Nevertheless, we can observe inflation and form a reasonably-informed view about whether the central bank is doing its job.

But what do macroprudential policies target? Typically it is to improve the positions of households during a recession, and potentially also to smooth the credit cycle. Where is the feedback? How can we know that it is achieving its results? The truth is that we cannot easily know. What we can see is that these tools have immediate distributional costs, making it difficult for lower-income and first-time-buyer households to access mortgages, for some uncertain benefits.

The objectives of macroprudential policy are more difficult to observe. Borrower-based tools are intended to limit the build-up of systemic risk, make households more resilient during a downturn, and, potentially, smooth the credit cycle. Authorities can observe household indebtedness as well as mortgage defaults. But these are intermediate outcomes. The ultimate benefit might be a recession in which fewer households default, fewer families are forced to cut their consumption sharply, and the fall in house prices is less severe that it would otherwise have been.

How do we know whether this has happened? As stated before, it is not easy to know. We may need to wait for a sufficiently large negative shock. Even when the shock arrives, typically several years later, we still need to estimate what would have happened without the policy. And while no severe downturn occurs, the benefits remain almost entirely counterfactual.

This produces a fundamental asymmetry. We can see many of the costs immediately. We can identify households that cannot obtain the mortgage they wanted, buy the house they wanted, etc. The benefits, by contrast, are delayed, contingent, and difficult to (empirically) attribute to the policy. A household that avoids default during a future recession may never know that a macroprudential intervention helped protect it; the same with an local market that sees a drop of house prices of 1% instead of the 2% that would have experienced without the macroprudential policy.

This does not mean that the benefits are not there. Nor does it mean that evaluating these policies is impossible. But it does mean that their evaluation requires very detailed data and careful empirical analysis. The necessary data tend to be collected by central banks. Therefore, central banks are often in the best position to do this evaluation. But this presents a problem: can we leave the assessment to the same institution designing and implementing the policy? The incentives might be distorted (Fabo et al., 2021 in the context of QE) and some central banks are cutting back on research. Independent scrutiny is therefore important not only for determining whether these tools work, but also for the democratic legitimacy of the powers exercised by central banks. Yet such scrutiny is possible only if outsiders can access the relevant data. This will be the subject of the fourth and final post.

Posted in Uncategorized | Tagged , , , , | 1 Comment

Borrower-based macroprudential tools (2/4): Distributional effects

Posted on July 5, 2026 by Francesc Rodríguez Tous

In the first post, I made the case for borrower-based tools: individually rational borrowing can be collectively excessive, and it can be optimal for the regulator to cap how much households can borrow against their homes. But I ended on a corollary: these tools affect some segments of the population disproportionally. This post explores some of the evidence of this unequal impact.

Leverage is not spread evenly across borrowers. If one looks at Figure 2 in our paper, reproduced below, we can see that both loan-to-value and loan-to-income ratios falling steadily with age and with income (particularly LTI). The people borrowing at the highest multiples are the young, the lower-income, the first-time buyers. For instance: the LTV in the first age quintile (the 20% younger borrowers) is around 80%, while for the oldest 20% of borrowers is around 40%. The same is true for the LTI: it goes from around 3.5 to below 2.5 for these same groups. This makes complete sense, it is nothing surprising. But it also means that limits on leverage will, by construction, affect these populations the most.

Figure 2 in Peydró et al. (2025).

So what is the evidence? Take the U.K., which is where our previous work comes in (Peydró et al., 2024). The Financial Policy Committee introduced in 2014 the following limit: no more than 15% of a lender’s new mortgages could carry a loan-to-income ratio of 4.5 or above; we call these “high-LTI mortgages”. Lenders that were already close to that ceiling cut their high-LTI lending, as intended. But they did something else, too: they made the high-LTI loans they did still grant bigger, steering them towards higher-income borrowers, so that lending to lower-income borrowers fell in aggregate. And the lenders who had room to increase high-LTI lending did not substitute this credit. In more competitive local markets, they actually raised interest rates on high-LTI loans to lower-income applicants—consistent with adverse selection concerns.

Ireland ran a similar experiment around the same time. Acharya et al. (2022) study the 2015 suite of LTV and LTI limits. Total mortgage issuance barely moved—but its composition tilted from lower- to higher-income borrowers, and house-price growth rotated away from “hot” (usually urban) towards “cool” (usually rural) counties. The banks most bound by the rules also went looking for risk elsewhere: more high-yield securities, riskier corporate lending. When you restrict the amount of risk banks can take in one part of their balance sheet, they typically go looking for it in another part.

Israel provides more evidence on the downstream effects of these tools, that is, how constrained borrowers change their behaviour after the introduction of these tools. Tzur-Ilan (2023) shows that constrained borrowers still bought homes, but they bought cheaper ones, further from the central business district, in neighbourhoods with worse socio-economic conditions and longer commutes. Note that while this paper does not explore further consequences, Chetty et al. (2016) shows that growing up in richer neighbourhoods has long-term effects in a series of outcomes.

There are many other case studies, which we review in the paper. But the previous three examples provide a coherent picture. More constrained borrowers, typically lower-income, younger, first-time buyers, find it more difficult to obtain a mortgage. This leads them to either not access the housing ladder right away or to go towards cheaper houses, usually in more rural or lower-income areas. This can lead to a cooling down of the house-price boom, which is one of the intended effects. Overall, these tools are effective but have distributional consequences.

But note that it is much easier to estimate the distributional “costs” of these tools (who does not get the mortgage) than the distributional “benefits” (who gets to keep their (cheaper) house once the recession hits). There is no doubt that avoiding deep financial crises benefits lower-income households the most. But putting a number on that is very difficult. We do this in Peydró et al. (2024) by using the Brexit referendum as a negative shock to the housing market. We find that more constrained areas had better price growth and lower defaults by lower-income borrowers. Yet for most countries, these negative shocks might take a long time to arrive, which can make the estimation of these counterfactuals really hard. And this has consequences for central banks and their policies. I will talk about this in the next post.

Posted in Uncategorized | Tagged , , | 2 Comments

Borrower-based macroprudential tools (1/4): what they are and why they exist

Posted on July 2, 2026 by Francesc Rodríguez Tous

I have a new chapter out with José-Luis Peydró, Jagdish Tripathy, and Arzu Uluc in the Research Handbook of Macroprudential Policy (Edward Elgar, edited by David Aikman and Prasanna Gai; the working paper version is here). It surveys what we have learned from more than a decade of borrower-based macroprudential tools—limits that restrict how much households can borrow, typically against their homes. There is a lot to say about these tools, so I am going to write a four-part series for the blog. But first things first. What does “macroprudential” mean?

The term contains two ideas: macro and prudential. Prudential regulation tries to ensure that banks and other financial intermediaries operate with sound levels of solvency and liquidity. It is not always successful, as the case of Silicon Valley Bank illustrates; but its objective is to keep the financial system robust. Historically, prudential regulation had been micro: it has focused on the soundness of individual institutions. The intuition was that if every single individual institution is sound, the system would be. This intuition is however wrong. And here is where the “macro” part comes in: it recognises that there are risks that can build up at system level that would be difficult to notice by looking at individual institutions. In other words, macroprudential policy understands that risks are endogenous and tries to prevent their excessive build-up.

What are some of these risks? Imagine that I am looking to buy a house. I want the best house for my family and myself, so I get as much mortgage as the bank will lend to me and buy the most expensive one I can afford. Great. The microprudential regulator looks at the mortgage market, realises that defaults are very low, and sees no issue with the bank providing the credit.

Now imagine a recession arrives. My household income suffers. But I still need to repay the mortgage. I might need to seriously cut other expenses. No dining out or ordering delivery. Maybe I was looking to change my car, but now I will have to wait. New clothes? A luxury. This all makes sense from my individual perspective. But what happens when many thousands of households do the same at the same time? It becomes a negative macroeconomic shock. Spending goes down, hiring too, maybe companies start firing employees… And the recession becomes deeper.

And nobody did anything wrong! When I got the mortgage, I weighed my own risks. I did not take into account how my spending cuts would affect aggregate demand. And why should I have? I am only one household; my effect on aggregate demand is minuscule. But as the number of affected households increases, the effect becomes significant. This is what economists call an externality: the private costs of my leverage (credit) are lower than the social ones, so we borrow too much. The consequences are well documented in Jordà, Schularick and Taylor (2016): recessions preceded by household credit booms are deeper and longer.

This is the case for macroprudential policies. A well-known one is the countercyclical capital buffer, or CCyB in regulatory jargon. Banks are regularly subject to capital regulation, which requires them to fund part of their assets with equity rather than only deposits and other debt. CCyB increases these requirements when the cycle is good, that is, when the economy is doing well and there is plenty of credit. The idea is to increase the solvency of banks precisely in a moment when doing so is cheaper. This may lead to less credit in the expansion, but this might be the right call given the negative externalities mentioned above. Moreover, better solvency in good times means better readiness for when the downturn comes. This is not just theoretical: there is evidence that the CCyB was useful in sustaining credit during the Covid-19 pandemic.

The CCyB, however, works on the lender’s balance sheet: it makes banks better able to absorb losses, but it does nothing about how much debt I take on in the first place. A second family of tools goes after the leverage itself: the so-called “borrower-based” tools. These are limits on how much households can borrow relative to the value of the property (loan-to-value) or to their income (loan-to-income), limits on the share of income going to repayments (debt-service), and tests of whether borrowers could keep paying if rates rose (affordability). They typically target mortgage lending, although some countries cover other types of credit as well. And they have been popular: the figure below shows how often countries have tightened them in recent decades. Note how direct this is. In the example above, the bank was happy to lend and the microprudential supervisor saw no problem; a borrower-based limit is the regulator saying no anyway—not because I could not repay, but because of what thousands of households like mine do to the economy when we all cut back at once.

Use of borrower-based macroprudential policies. Net number of borrower-based measures (LTV and DSTI limits) tightened each year (tightening minus loosening actions, summed across all countries). Source: IMF iMaPP Database (Alam et al., 2019).

That directness comes with a catch, and it is the theme of the rest of this series. The CCyB binds on banks; borrower-based tools bind on households. And not on all households equally, because not everyone needs to borrow at high multiples to buy a home: the young, the lower-income, the first-time buyer do—households with the income to service a mortgage, but without the savings for a large deposit. These tools are distributional by design. And this is the topic for the next post.

Posted in Uncategorized | Tagged , , | 3 Comments

South Korea

Posted on June 23, 2026 by Francesc Rodríguez Tous

I find South Korea a fascinating country. One of the economic miracles of the second half of the 20th century, South Korea not only went from extremely poor in the 1950s, when it suffered the invasion of North Korea and the subsequent war, to sitting among high-income countries nowadays, but also became a cultural powerhouse, dominating the music and film scene in the region. Yet South Korea is ending: its fertility rate is the lowest in the world, and absent any radical changes, population will shrink from around 52 million today to somewhere around 20 million by the end of the century.

The economic story is staggering. In 1960, South Korea’s GDP per capital was just over $1,000 in today’s prices, among the poorest countries in the world, behind much of sub-Saharan Africa. By 2024, it stood at over $37,000: a 36-fold increase in 64 years or, put differently, a 5.75% annual growth on average for the whole period. An authentic economic miracle.

But there is more. 한류 (Hallyu), or “Korean Wave”, refers to the global popularity gained by Korean pop culture, from videogames, to cinema, music, products, and more. This started taking off during the 2000s, first to close-by regions, such as Japan and China, and then globally in the 2010s. Shows such as Squid Games, movies such as Parasite, and K-Pop bands such as BTS and Blackpink have attracted cult followings all around the world. South Korea has a prestige and attention way beyond the size of its economy.

So why is it ending?

Demographers say that a society needs a fertility rate of at least 2.1 (children per woman) to maintain a stable population without immigration. South Korea crossed below that threshold in 1983 (before I was born!) and never came back. By 2023, the rate reached 0.72, the lowest ever recorded for any country (with the exception, I hope, of the Vatican). It rose slightly in the following year but nothing to prevent the overall trend: from 6 children per woman in 1960 to 0.75 in 2024. While most countries around the world are experiencing declining fertility rates, the South Korean evolution beats everyone else.

The consequences are as clear as basically inevitable. The population of South Korea will continue declining and this decline will accelerate in the second half of this century, finally halving as we approach 2100. An unprecedented level of immigration could help slow down the fall, but this would need to be truly a massive program and it would probably only push the fall towards the 22nd century. The same is true for any pro-natalist policies: some have already been tried by the South Korean government, with marginal effectiveness given the figure above. Any intervention that can move the needle on the population projections would need to be massive in scale. And it would only move the needle slightly.

What is the problem of halving population, if it was at a similar point in the 1950? Couldn’t it simply start growing again and follow the same cycle? Unfortunately, there is a crucial difference: the age composition of the population. The median age in 1950 was under 20 years: as the figure below shows, there were almost as many children as working-age people. This will not be the situation in 2100: around half the population in South Korea will be above 60 years old and the “elderly” (65 and above) will be as many as the working age population by the end of this century. This is older than any country in the world today (except, again, probably the Vatican). Not what we would describe as a dynamic fast-growing society.

What does this all mean? South Korea is the most extreme example of a problem that is happening in most, if not all, richer countries: younger generations are deciding not to have children. Exactly why is the subject of academic as well as political and popular debate. I have absolutely nothing to bring to this discussion. This post is just to explain why South Korea is at its absolute peak right now and will most likely never be as influential as it is today. It is a wonderful country to visit; I did so a couple of years ago (see some pictures below!) and hope to do it again soon.

Seoul at night, view from going up to Seoul Tower
My daughter skipping stones in the Cheonggyecheon (청계천), Seoul.
My daughter dressed in traditional hanbok (한복), in Gyeongbokgung (경복궁), Seoul.

Posted in Uncategorized | Tagged | Leave a comment

Who leaves when a central bank shrinks?

Posted on June 14, 2026 by Francesc Rodríguez Tous

A couple of months ago, I started noticing something on LinkedIn: economists I knew at the Bank of England were announcing they were leaving the institution (always, curiously, with a picture of the building behind them). Many of them very good researchers. First it was just one, then two, but by now I have honestly lost count. It turns out that this trend is a consequence of the voluntary exit scheme that the Bank implemented and that closed in mid-January.

The numbers are substantial. Though not official, the reported numbers of staff leaving are around 450, which represents 8% or so of the total workforce. Apparently, more than 700 employees volunteered to leave the Bank, so not everyone was allowed to do so under those conditions (reporting here). It says something when you open a voluntary redundancy scheme and it becomes oversubscribed by over 50%.

But why is the Bank doing this? This move responds to budget concerns that coincide with the implementation of the Bernanke Review. The Review provided a set of recommendations, and can be read here. One set of recommendations that apparently has hit the budget substantially is the upgrading of the infrastructure of the forecasting. While forecasting errors at the Bank of England were not particularly bad—compared to other central banks—Bernanke found that the software infrastructure supporting the process was lacking. This led the Bank to act to improve it.

As it sometimes happens with these reports, full of (very relevant) recommendations, the leadership decides to focus on those that they prefer. For instance, under recommendation 6, Bernanke states that “Researchers with doctorates should continue to spend part of their time in undirected or loosely directed research, with the best researchers afforded the opportunity to continue working on their individual agendas throughout their careers.” Given how many PhD researchers I have seen (on LinkedIn) leaving the Bank, this recommendation has not been particularly prioritized.

But trimming a policy institution does not need to be a bad thing. The issue is, however, who leaves. Leaving is a decision that is based on the stage of one’s career, immediate short-term pressures, and outside options. Other things equal, those with better outside options will tend to take the leave. Note that not everyone was allowed to get the redundancy package. But judging by the number of LinkedIn posts of top researchers leaving the Bank, it is clear that they were not the profiles that they wanted to keep. So those with better outside options are more likely to apply for the package, which means that the research quality of the institution is likely to drop. This is the opposite of what you would want: a well-designed scheme would retain the best, not lose them.

But why should the Bank of England, or any other policy institution, have PhD economists doing research? Let’s put some numbers for context. The Bank of England lists 80 researchers on its website, and this might not be updated with the recent departures. So we are talking about a small percentage of the overall workforce. And having rigorous experts that can lead the work on understanding the effects of policies and other actions is crucial for improving the functioning of the institution. This is true in general, but it might be specifically important now with new technologies with uncertain effects on the labour market, a push towards de-regulating the financial sector, and a strong pressure to make central banks less independent.

I also think that the impact might be uneven. On the monetary policy front, we know that the Bank is working on the forecasting part, so the loss of the research talent might be compensated by that. But other areas, such as financial stability or more policy-oriented departments, the departures might be felt more acutely. And central banks have a really thankless job on the financial stability side: if regulation works well, the calls for de-regulation come. If regulation does not work and risks materialize, then of course the central bank is to blame for it. Losing academic and analytical rigour in this moment can end up being a liability later. Let’s hope it does not come to that.

Posted in Uncategorized | Tagged , , | 2 Comments

Generalists in the Age of AI

Posted on June 7, 2026 by Francesc Rodríguez Tous

For most of modern intellectual history, the generalist has been losing. The trend is visible everywhere, including in academia. At my university, Bayes Business School, the finance faculty is divided into groups: banking, asset pricing, M&A, real estate,… These are sub-fields of finance, which is a sub-field of economics, which is a sub-field of the social sciences. And within each group, researchers specialise further still. While this fragmentation is not completely arbitrary, I would argue parts of it are. Disciplines, or sub-sub-(sub-sub…)fields, develop for many different reasons, sometimes by chance: because a set of questions does not fit comfortably inside the mainstream, you get, for example, behavioural economics. Whatever the reasons for this fragmentation, the 18th-century ideal of the “natural philosopher,” equally at home in mathematics, physics, biology, and philosophy, became not just rare but structurally disadvantaged. Being a generalist became increasingly expensive: as each field grew deeper and more technical, the cost of being credible across multiple of them grew beyond what most people could afford.

Large language models are changing this in two ways. First, they lower the cost of acquiring breadth. Learning the basics of a new field (or sub-field, or… you get my point) used to take months of careful reading. But now you can learn its key concepts, frameworks, debates, etc. in a few hours of good conversation with an LLM. Not expertise, of course, but orientation. And orientation matters. The generalist’s traditional handicap is shrinking fast, really fast. Second, and less obviously, LLMs raise the return to genuine cross-domain thinking. This is because of what LLMs are not particularly good at: knowing when a connection between fields is actually deep. They are very good at connections that are already legible in language. They excel at the kind of analogy that is already there in the text. What they miss are the abstract isomorphisms—when an insight from one field reframes a problem in another. Kahneman and Tversky are a good example: the insight that human decision-making under uncertainty follows systematic, predictable patterns was a psychological observation that turned out to be one of the most important ideas in economics of the last fifty years. The fields were asking closely related questions—how do people choose?—in different languages. The LLM makes the cheap connections; the scarce thing is the judgment to know which connections are deep and which are superficial.

None of this is an argument against specialisation. The research frontier in any serious field will continue to be pushed by people with deep, compounding expertise—the kind of calibrated pattern recognition that comes from years of immersion and cannot be shortcut. If anything, LLMs will help specialists too, handling the retrievable parts of their work and freeing them for the genuinely hard problems. The point is not that generalists replace specialists, but that the relative return to breadth is increasing for the first time in a long while.

We may be entering a moment that, in a small way, resembles the age of the natural philosophers—not because any individual can master all of science, which is impossible, but because the person who has thought seriously across multiple fields, and who can spot the connections that no algorithm will surface, is suddenly more valuable than they have been for centuries. The cost of being a generalist is falling. The value of being a genuinely good one is rising. That is an unusual combination. How long will it last?

Posted in Uncategorized | Leave a comment

The Quantum Bomb Experiment

Posted on January 10, 2023 by Francesc Rodríguez Tous

I have been reading about the history of quantum theory and how this history has shaped the different interpretations of it. ‘Helgoland’, by Carlo Rovelli, ‘Quantum’, by Manjit Kumar, and ‘What is real?’, by Adam Becker, are excellent books that discuss these issues. The weirdness of quantum physics is fascinating. And an experiment that illustrates this weirdness wonderfully is the quantum bomb experiment.

Imagine that you have several bombs that are activated—i.e., they explore—when a photon (the particle that light is made of) passes through them. You need to get a bomb for a mission, but some of the bombs do not work anymore: their photon detector is broken. The only way to check whether a bomb works is to send a photon and see what happens. Can you find a bomb that works without actually make it explode?

It turns out that using quantum physics you can. Intuitively, when the bomb works, it is a detector to observe where the photon is, so it forces the collapse of the wave function; but when it does not work, quantum superposition is maintained. Using this difference, one can construct a device that can tell you whether the bomb works even when the photon does not go through it. I am not going to explain the inner workings of this device, but I’ll leave a couple of videos that do so.

The Bomb Experiment, by Sabine Hossenfelder:

Elitzur-Vaidman Bombs, by Barton Zwiebach:

Posted in Uncategorized | Tagged , | Leave a comment