The 2016 Sterling Flash Episode

From The Bankunderground

In the early hours of the morning of 7 October 2016, the sterling-US dollar exchange rate fell by nearly 10% within around 40 seconds. Most of this movement was reversed within the ten minutes that followed. This was one of a series of such ‘flash’ episodes in major financial markets – that is sharp and short-lived movements in price, which vastly exceed perceived changes in economic fundamentals. It certainly didn’t fail to catch the eye of policymakers, or the media.

This post summarises a recent working paper by Bank staff to understand what happened and why.

What happened?

To get at this question, we took high frequency data from the Thomson Reuters platform. Day-to-day, this platform facilitates between five and ten per cent of trading in the sterling-US dollar spot market – one of the most liquid currency pairs in the world.

Chart 1 shows data taken from this platform around the episode:

The triangles show the prices at which individual transactions took place. Those in blue (pointing down) indicate transactions initiated by a participant seeking to sell sterling. Those in green (pointing up) indicate those initiated by an order to buy.

The shaded regions show the cumulative distribution of limit orders around these prices. Limit orders are unexecuted orders to buy or sell sterling posted by prospective traders.

The relative weight of the shading on the chart shows the quantity of limit orders between a given price and the limit orders to buy/sell at the highest/lowest prices (the ‘best bid/ask prices’). As might be expected, prices further from the best bid/ask are shaded in darker colours. This indicates that there lies a larger cumulative quantity of limit orders between them and the best bid/ask.

The black line shows the midpoint – or ‘mid-price’ – between the best bid/ask prices.

From this chart we can construct a rough narrative of events:

In the minute preceding the crash, between six and seven minutes past midnight (00:06:00 and 00:07:00 British Summer Time), there was quite a large depth of orders both to buy and sell sterling (£60 million of orders in the observed ten levels of price closest to the best bid and ask prices).

But at around 00:07:00, an imbalance started to develop – with the quantity of orders to sell sterling starting to exceed those to buy.

It was at this point that a rapid succession of trades took place in sterling, at rapidly declining prices.

This imbalance became particularly severe around 00:07:17 BST. This can be seen from the large white areas in the graph, which indicate there to be (close to) no orders to buy sterling. In the half minute that followed, market functioning was severely impaired, with large ‘gaps’ in price visible between trades.

The quantity of, and balance between, limit orders to buy/sell sterling recovered after about 30 seconds but deteriorated severely again after around one minute. Shortly after 00:09 BST, there was a further sharp reduction in orders to buy sterling (corresponding to another white area on the chart).

The order book started to increase in depth around 00:09:30 BST, around 150 seconds after the initial sharp movement in price.

Thankfully, the events of the night of 7 October 2016 were without lasting consequences for financial stability, or the integrity of the functioning of the market for sterling.  Higher than usual volumes were observed during the day that followed, and measures of illiquidity (including bid-ask spreads) remained slightly elevated; but broader spill overs were generally limited.

That, however – understandably – hasn’t stopped the search for answers.

…and why?

In its report on the episode, the Bank for International Settlements (2017) found the movement in the currency pair to have resulted from a confluence of factors. These included larger-than-normal trading (predominantly selling) volumes at a typically illiquid part of the trading day. There were also sales of sterling by some market participants seeking to limit the risk associated with their positions in options markets, and to execute client orders in response to the initial fall in the exchange rate.

One important outstanding question is the degree to which the change in price witnessed during the episode was in line with the imbalance between observed orders to buy and sell.

We’d expected an imbalance in order flow during an episode like this to increase the change in price that results from a trade (or trades) of a given size. An imbalance in orders might be interpreted as indicating that some market participants were party to superior, or more up-to-date, information. This might, in turn, widen the spread at which other participants were willing to buy/sell sterling.

To assess this empirically, we develop an estimate of the change in price that is likely to result from an imbalance in orders in foreign exchange markets. This is based on previous literature. It is robust to the possibility that a large order might be split up into a series of smaller orders. This is important, because such splitting of large orders is common in foreign exchange markets, because, by transacting in smaller size, market participants can obtain a better price.

The blue bar in Chart 2 shows the range of estimates of change in price given by this model, when calibrated to past movements in the sterling dollar exchange rate.  These imply that the observed orders to sell sterling during the flash episode are consistent with a decrease in the sterling-US dollar exchange rate of between 1.03% and 2.87%, depending on the precise choice of parameters.

Chart 2: The (in)consistency between observed changes in price and those expected given observed orders to buy/sell sterling

The dots in the right-hand columns compare these estimates with the observed decline in the exchange rate. The purple triangle shows that which took place early in the episode, between 00:07:00 and 00:07:15 (roughly step (2), above). The yellow square shows the peak-to-trough fall in sterling over the entirety of the episode.

From this we can see:

  • The initial fall in sterling, during the early part of the episode, of 1.8%, is consistent with the range of estimates based on the observed imbalance of orders. This suggests that the movement in price was consistent with the arrival of a large order to sell sterling.
  • But the larger change in price that occurred between 00:07:00 to 00:07:17 cannot be explained by expected price impact of trades alone.

Was there another factor at play?

That the overall fall in sterling so vastly exceeds that predicted by our model might suggest that some some other driver might have been at play.

The report by the BIS suggests a number of other factors that might have played a role in reducing available liquidity during the episode. These include the temporary withdrawal of some market participants from their role as market makers. This dynamic may have, in part, reflected the presence of staff with lower risk limits and appetite at some institutions at that time of day. An automatic pause in trading in sterling futures contracts may also have led to a reduction of liquidity in the cash market, because some market makers are thought to rely on futures as a guide to the price at which they offer to buy/sell currency in cash markets. It is, however, difficult to know what weight to place on these different explanations.

All else equal, such a reduction in liquidity would have increased the resulting fall in price beyond that estimated to be in line with observed trading volume.


The events of 7 October 2016 represented one of a series of flash events occurring in electronically traded markets. No such events have, as yet, had longer lasting consequences for market functioning or stability.

Nonetheless, policymakers have recently pointed to the clear onus on central banks and the regulatory community to understand developments in these markets, and how they behave during periods of stress.

This work represents one step in that effort.

Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees

Crypto Is Not The Future Of Money

Crypto-currencies do not stand up as a new form of money says Mark Carney, Governor of the Bank of England, speaking on “The Future of Money“. That said, the underlying technologies and capabilities, have potential.

The long, charitable answer is that crypto-currencies act as money, at best, only for some people and to a limited extent, and even then only in parallel with the traditional currencies of the users. The short answer is they are failing.

They are poor stores of value, an inefficient media of exchange and are virtually non-existent units of account.

Authorities need to decide whether to isolate, regulate or integrate crypto-assets and their associated activities.

This is probably the strongest statement on the subject so far from a Central Banker.

But, whatever the merits of crypto-currencies as money, authorities should be careful not to stifle innovations which could in the future improve financial stability; support more innovative, efficient and reliable payment services as well as have wider applications.

The underlying technologies and capabilities, have potential, given the right regulatory frameworks.

Their core technology is already having an impact. Bringing crypto-assets into the regulatory tent could potentially catalyse innovations to serve the public better. Indeed, crypto-assets help point the way to the future of money in three respects:

Decentralised peer-to-peer interactions

Crypto-assets are part of a broader reorganisation of the economy and society into a series of distributed peer-to-peer connections across powerful networks.28 People are increasingly forming connections directly, instantaneously and openly, and this is revolutionising how they consume, work, and communicate.

Yet the financial system continues to be arranged around a series of hubs and spokes like banks and payments, clearing and settlement systems. Crypto-assets are an attempt to create the financial architecture for peer-to-peer transactions. Even if the current generation is not the answer, it is throwing down the gauntlet to the existing payment systems. These must now evolve to meet the demands of fully reliable, real-time, distributed transactions.

Underlying technologies offer to transform the efficiency, reliability and flexibility of payments.

The technologies underlying crypto-assets, particularly distributed ledger, can:

  • Increase the efficiency of managing data;
  • Improve resilience by eliminating central points of failure, as multiple parties will share replicated data and functionality;
  • Enhance transparency (and auditability) through the creation of instant, permanent and immutable records of transactions; and
  • Expand the use of straight-through processes, including with “smart contracts” that on receipt of new information, automatically update and if appropriate, pay.

These properties mean distributed ledger technology could transform everything from how people manage of their interactions with public agencies, including their tax and medical records, through to how businesses manage their supply chains.

A Central bank digital currency (CBDC) accessible to all.

Crypto-assets raise the obvious question about whether their infrastructure could be combined with the trust inherent in existing fiat currencies to create a central bank digital currency (CBDC).

Currently only banks can hold central bank money electronically in the form of a settlement account at the Bank of England. To be truly transformative a general purpose CBDC would open access to individuals and firms.

The Bank has an open mind about the eventual development of a CBDC and an active research programme dedicated to it. That said, given current technological shortcomings in distributed ledger technologies and the risks with offering central bank accounts for all, a true, widely available reliable CBDC does not appear to be a near-term prospect.

Moreover whether it is desirable depends on the answers to a series of big policy questions. While these are largely for another speech, I will note that a general purpose CBDC could mean a much greater role for central banks in the financial system. Central banks may find themselves disintermediating commercial banks in normal times and running the risk of destabilising flights to quality in times of stress.

There are also broader societal questions (that others would need to answer) such as how society balances privacy rights with the extent to which the information in a CBDC could be used to fight terrorism and economic crime.

Peer to Peer – Scale and Scalability

From BankUnderground.

Peer to Peer (P2P) lending is a hot topic at Fintech events and has received a lot of attention from academia, journalists, various international bodies and regulators.  Following the Financial Crisis, P2P platforms saw an opportunity to fill a gap in the market by offering finance to customers and businesses struggling to get loans from banks.  Whilst some argue they will one day revolutionise the whole banking landscape, many platforms have not yet turned a profit.  So before asking if they are the future, we should first ask if they have a future at all. Problems such as a higher cost of funds, or limited ability to scale the business, may mean the only viable path is to become more like traditional banks.

Present Scale and Profitability

P2P activity has now been around for over a decade. The fastest growth has been in China, followed by the USA and the UK (total new alternative finance provision in 2016: China – $243bn,US – $35bn, UK – £4.6bn). P2P lending platforms offer an online marketplace and depend on both the external supply of investment and the demand for loans. Currently, platforms lack product diversification with their revenue deriving from origination and servicing fees. As such, platforms are extremely reliant on continuously attracting and matching loans for investors and borrowers.

Despite having substantial lending volumes, many big UK and US platforms are still making operating losses despite their rapid growth.

In the following two sections we explore two ways in which P2P platforms could make money: first, by scaling up their existing business models; and second, by changing their business models altogether.


It may be that some platforms have made a conscious decision to favour growth over profitability for now, with a view to realising economies of scale. Central to this business model is i) whether there is sufficient appetite for the P2P investments and loans for the platform to reach scale and ii) whether their revenues can exceed the costs, even if they achieve higher scale.

Appetite for P2P products

P2P lending is still relatively young in the UK. As matchmakers, P2P lending platforms need to keep attracting new customers from both sides of the equation in order to grow. This is not a straightforward task: for example, supply of funds might be available but there might be lack of quality borrowers.  Or alternatively, they might have a slew of willing borrowers but are unable to tempt sufficient investors to finances them. A slowdown on either side affects platforms’ growth.

In the UK, P2P lending to businesses (mostly small to medium sized) is more substantial than to consumers. By contrast, in China and the US, P2P lending is mostly consumer focused.  On the supply side, retail investors have dominated the market, but the role of institutional investors has been increasing.

In the past few years, the market in the UK has been growing very rapidly, with year on year new lending growth in the region of 100%. But new data from the Cambridge Centre for Alternative Finance suggests that although activity is still growing, the pace of that growth has slowed down considerably.  In 2016, total new lending grew by roughly 50% in the UK and 20% in the US.

This, coupled with indications from some lenders that they are running into borrower constraints (i.e. their lending activity is being restricted by the low amount of potential borrowers), suggests that the traditional P2P lending model, that is widely used in the UK, may be reaching its limits. If growth rates continue to slow, platforms will find it more difficult to achieve the size necessary to fully realise their economies of scale.

One way that individual platforms may attempt to tackle this issue is through consolidation but given that the industry appears to be quite concentrated already (currently there are 3-4 major platforms that are key players), there might be only limited room for further concentration.  At the end, it may be that sustainable profitability may only be achievable with a few large lenders on the market.  Parallels can be drawn with other tech industries, where initially there were a number of players, but they gradually consolidated down to one or two market leaders (e.g. Google, Facebook).

Cost Structure

However, it might be the case that even following consolidation and hence larger scale, platforms still will not make money. For example, Uber is still unprofitable despite being a market leader on a large scale. And the two largest lendingplatforms in the US, who are dominant providers of P2P loans and several times larger than UK platforms, are also loss making.

The answer may lie in examining platforms’ cost structure more closely.  In theory, P2P platforms have operating cost advantages; they have no legacy costs, no requirement for branch networks and lower regulatory costs. At face value, these should be lower than for traditional banks. And, unlike traditional banks, these should be largely unrelated to scale- because, like other tech disruptors, their main cost is setting up and operating a platform. That opens up the possibility of undercutting traditional banks if platforms can achieve high enough lending volumes to overcome their fixed costs and realise these cost advantages.

But we must also think about the P2P lenders’ cost of funding- i.e. the interest rate they have to offer lenders to induce them to invest.  If this cost of funding is sufficiently higher, this could undo the advantage of lower operational costs. In reality, banks are able to borrow money at a lower cost, so even if P2P lenders have the slimmer cost base they may not be able to undercut banks, despite offering higher interest rates to borrowers.  Just scaling up might not be enough and some platforms may need to adjust their business model altogether…

Becoming a bank-like P2P platform

In the UK, P2P lending platforms have generally begun with ‘traditional’ or ‘pure P2P’ business models. But what started as pure and simple has been continuously evolving. Platforms have already been experimenting with new business models and techniques.

Some “P2P” platforms have had success operating a ‘balance sheet’ lending model. These platforms’ business model is not pure peer-to-peer lending, because the bank itself co-invest with investors, putting their ‘skin in the game’.  To do this successfully, such platforms tend to also concentrate on a particular lending market (e.g. property lending).

Other platforms might turn to traditional banking (one of the leading UK platforms is already applying for a banking licence) perhaps to diversify range of services they offer. For example, platforms will be able to offer FSCS-protected deposit accounts for savers and personal loans, car finance, and credit cards for borrowers, alongside their P2P products. Another reason is that FSCS protected deposits mean lower funding costs. Guaranteed deposits also mean that platforms can be listed on “best buy” comparison portals for savings account, creating a new potential market to tap for funds.

A natural question that follows – is there a fundamental difference for customers between banks and P2P platforms? The answer is not straightforward. Undoubtedly, platforms offer some distinct benefits for investors compared to banks– an opportunity to lend directly to businesses and retail consumers with relatively small amounts of investment, and achieve a higher rate of interest than is available from traditional savings accounts. On the other hand, banks offer deposits that are covered by the FSCS and so are less risky to P2P investors (although less risk averse investors might find that a better place to be on the risk-return trade-off).

On the borrowers’ side, there may be less differentiation. Our internal analysis (Chart 2) shows that interest rates on personal loans arranged via P2P platforms are competitive but not significantly lower than the rates available from banks. The exception is low-value loans (i.e. around £1-2k), where banks’ manual processes and fixed costs, make it uneconomical for them to compete.

But P2P platforms are not the only ones adapting. As P2P lenders start to become more like banks, banks are starting to become more like P2P lenders in some respects.  To counter the possible competitive threat from P2P lenders, banks have started to offer quicker and more user-friendly loan applications services (including quick, all-digital SME lending services). For a potential borrower, the difference between a P2P platform and a bank becomes less obvious. So, to stay in the game, platforms will need to compete for borrowers’ attention by offering a wider range of bank-like services.

Ironically, it might be the case that as much as the platforms have wanted to disrupt the banking model, they might need to turn towards it to grow and to achieve profitability.


Ultimately, the feasibility of scaling up depends on the balance of the two factors: a continuous appetite for P2P investments and loans, and whether the revenues can be higher than the costs when they do scale up.

The answer could be that platforms will need to consolidate and adjust their business models if they wish to have a significant and lasting presence in the financial system. A key part of this may be turning to banking: whether via partnering with banks or by offering bank-like services themselves. Peer-to-peer lending might be changing the world, but perhaps it will have to change itself first.

Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.

The rise and fall of interest only mortgages

From Bank Underground.

The interest-only product has undergone tremendous evolution, from its mass-market glory days in the run-up to the crisis, to its rebirth as a niche product. However, since reaching a low-point in 2016, the interest-only market is starting to show signs of life again as lenders re-enter the market.

The chart shows how in 2006, interest-only mortgages were used by borrowers in the UK to purchase a higher value property than they otherwise might have been able to afford with a capital or repayment mortgage. This was because monthly repayments were lower for the interest only mortgage, and relatively high Loan-to-value (LTV) ratios meant that a lower deposit was required.

The drawback was that borrowers needed enough funds to repay the entire capital outstanding at the end of the mortgage term.

Since the crisis, rules have come in requiring lenders to have credible repayment strategies for the capital outstanding, and implement stricter underwriting standards on affordability.

Accordingly, the interest-only product has evolved into a much more niche product (falling from 42% of new lending in 2007 to just 7% of lending in 2016), predominantly targeted at higher-income borrowers. The combination of shorter mortgage terms and low LTVs suggest that borrowers are now using the product as a source of cheap borrowing for other purposes, rather than solely for house purchase.

However more recently, there are signs that lenders are starting to expand interest-only lending again, which rose to £5.4bn in Q3 2017, a 45% increase on the previous year.

Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.


Population Ageing and the Macroeconomy

From The Bankunderground.

An unprecedented ageing process is unfolding in industrialised economies. The share of the population over 65 has gone from 8% in 1950 to almost 20% in 2015, and is projected to keep rising. What are the macroeconomic implications of this change? What should we expect in the coming years? In a recent staff working paper, we link population ageing to several key economic trends over the last half century: the decline in real interest rates, the rise in house prices and household debt, and the pattern of foreign asset holdings among advanced economies. The effects of demographic change are not expected to reverse so long as longevity, and in particular the average time spent in retirement, remains high.

An unprecedented demographic change…

Population ageing is typically summarised by the old age dependency ratio, the ratio of the population over 65 relative to the working population. In industrialised countries, this ratio has risen from under 15% in 1950 to over 30% in 2015, and is forecast to rise to 50% in the next 20 years (dark blue line in Chart 1). Looking at a few countries where population data is available from the 19th century, shown in the dashed lines in Chart 1, this trend is unprecedented. While falling birth rates do play a part, the trend is driven predominantly by increasing longevity. The same ageing process is also happening, albeit more slowly, in middle and low income countries.

Chart 1: Old Age Dependency Ratio across industrialised countries (%)

Note: The ratio is defined as population over 65 divided by population aged 20-64. The thick navy line shows aggregate data from the UN Population Statistics for 17 industrialised countries (Australia, Austria, Belgium, Canada, Denmark, France, Germany, Ireland, Italy, Japan, Netherlands, New Zealand, Portugal, Spain, Switzerland, UK and USA), and dashed navy lines show the high- and low-fertility scenarios in their projections. Thin dashed lines show the historical data for Belgium, Denmark, France, Netherlands, Switzerland and UK from The Human Mortality Database.

From an individual perspective, this trend reflects an increasing fraction of life spent in old age. People that reach retirement age can now look forward to living a further 20 years, on average. Assuming no change in the retirement age, this number is projected to rise to almost 30 years for generations born 20 to 30 years from now. Consider that when Bismarck, Lloyd George and others pioneered state pensions over a century ago, workers were lucky merely to reach retirement age. While raising the retirement age can work to stabilise the time spent in retirement relative to time spent working, the increases proposed in most advanced economies  are not yet enough to offset this rise in life expectancy.

… with a profound effect on desired wealth accumulation…

One of the primary ways that population ageing affects the economy is through savings and wealth accumulation. While many developed countries have some form of implicit transfers from workers to retirees through state pensions, private saving remains an important component of retirement income. This is evident in the life-cycle pattern of wealth holdings: in the United States, where the most data is available, wealth peaks close to retirement age and then falls gradually (Chart 2).

Chart 2: Net worth over the life-cycle, with and without housing (thousand US Dollars)

Note: The data is taken from Survey of Consumer Finances, averaged over 1989-2013. The dark solid line shows total net worth, and the light dashed line shows net worth excluding housing wealth.

The rise in life expectancy raises the economy’s desired level of wealth for two reasons. Firstly, people need to accumulate more wealth during their working life to fund their consumption over a longer expected retirement. In terms of the life-cycle profile of wealth, all else equal, this would mean that wealth rises more steeply and reaches a higher peak. Secondly, even without any change in behaviour over the life-cycle, the changing population age structure would imply rising aggregate wealth. Specifically, Chart 2 shows us that households accumulate much of their wealth by around age 50, and hold on to that wealth throughout retirement. Therefore, when adding up individuals’ wealth, the increasing share of people in these high-wealth stages of life will imply a higher aggregate level of wealth.

… With far-reaching macroeconomic implications…

What are the macroeconomic effects of this rise in wealth? In terms of a simple concrete example, household savings find their way to firms, who invest them in machines, structures and intangibles such as branding and research. To the extent that it is harder for firms to employ extra machines as profitably as existing ones, i.e. that the returns to these investments are diminishing at the margin, households get lower returns as their savings increase. In other words market interest rates are lower.

The effects of ageing do not stop there. As desired wealth rises, the demand for other assets, including for example housing, also rises, pushing up on prices. Borrowing also rises, in response to both the fall in the interest rate, which makes borrowing cheaper, and the rise in house prices, as younger and less wealthy households borrow to buy housing. Within open economies, countries that are ageing relatively faster will accumulate assets in countries that are ageing relatively more slowly. This would explain why rapidly ageing countries, like Japan or Germany, are lending money to relatively younger countries, such as Australia.

While the mechanism described above is, of course, simplified, there is evidence that the ratio of capital-to-GDP is indeed much higher now than in the past (Chart 3). This provides additional evidence in favour of our underlying mechanism.

Chart 3: Measures of Capital-to-GDP in industrialised countries (Index)

Note: The blue line shows an index of the ratio of the capital stock to value added in the US business sector from Fernald (2014), equal to 1 in 1947. The red line shows an index of the ratio of capital services to GDP for 19 OECD countries, which, for comparison, we normalise to equal the Fernald measure in 1985 when the series begins.

… which are quantitatively significant

In a recent staff working paper, we have embedded the mechanisms described above in a life-cycle model, in order to quantify these effects for industrialised countries. Households in our model follow the life-cycle patterns of work, home ownership and wealth that we see in the data. This means we take as given that retirees keep their high level of wealth, and assume that future retirees will do the same. We assume that households know their life expectancies accurately, take account of current and expected house prices and interest rates, and hence plan ahead for their future consumption. This implies that households are able to save more in anticipation of their longer retirements, rather than having to work longer or give up their consumption in old age.

In equilibrium, the real interest rate adjusts to balance the supply of capital from the household with the demand from firms. This sets to one side the fact that, in practice, there is a large spread between the risk-free interest rate in financial markets and the interest that firms earn on new investments. This spread is comprised, among other things, of various premia for liquidity and risk, and profits that firms earn in excess of their costs, which we abstract from. House prices also adjust to balance demand for housing with a supply that we assume is fixed per head.

We allow birth and death rates to fall in line with UN data and projections for industrialised countries, as defined in Chart 1. This will be the only external driver of the dynamics in our model. We set the model to match the level of interest rates, and some other aggregate variables in the 1970s, and then measure how the demographic trends have affected the economy since then.

We find that demographic change alone can explain 160bps of the 210bps decline in interest rates since the early 1980s measured by Holston et al. (2017). The model predicts an increase in house prices of over 45% since 1970, corresponding to around 75% of the change observed in the data (Chart 4a). It also explains the doubling of the private debt-to-GDP ratio between 1970 and the start of financial crisis, an increase of around 45pp (Chart 4b). These results confirm that ageing does have a sizeable economic impact. Implications for cross-country imbalances are also important: the pattern of foreign asset accumulation (net borrowing and lending) across industrialised economies predicted by the model explains almost 30% of the variation observed in 2010.

Chart 4: House prices and gross private debt-to-GDP in the model and in the data


Note: House prices are shown in percentage deviation from the 1970 value, and private debt-to-GDP is shown as a percentage. In both cases, the blue solid lines show the results of our model simulations and red dashed lines show the equivalent series in the data.

Finally, the model allows us to gauge the effects of demographics going forward, based on the projections in the UN data. Importantly, these effects are set to increase over the future. The retirement of baby boomers is sometimes cited as potentially driving a decrease in aggregate savings. This does not happen in our quantitative model, both because new generations are expecting to live ever longer into retirement, and are therefore saving more, and because we can expect that baby boomers will retain their high wealth levels throughout retirement. This means that population ageing will continue to keep long run interest rates lower than they would otherwise be, as long as life expectancy, in particular the expected duration of retirement, remains high.

Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.

Bank Profits and the Yield Curve

From The Bank Underground.

The conventional wisdom amongst financial market observers, academics, and journalists is that a steeper yield curve should be good news for bank profitability.   The argument goes that because banks borrow short and lend long, a steeper yield curve would raise the wedge between rates paid on liabilities and received on assets – the so-called “net interest margin” (or NIM).  In this post, we present cross-country evidence that challenges this view.  Our results suggest that it is the level of long-term interest rates, rather than the slope of the yield curve, that drives banks’ NIMs.

Net interest margins are calculated as the interest banks earn on their assets—e.g. on the loans they make — minus the interest they pay out on their liabilities – e.g. the interest they pay to savers.  Meanwhile, the slope of the yield curve is defined as the difference between the long-term interest rate (10 year government bond) and a short-term rate.

The conventional wisdom follows from  banks’ fundamental business model— to act as maturity transformers by borrowing short term (e.g. from deposit accounts) and lending long term (e.g. through mortgages or loans to companies).  This activity is typically profitable as short-term interest rates are usually lower than long- term interest rates. This reflects the fact that depositors are generally willing to sacrifice returns because they value the liquidity of holding their money in cash rather than in an illiquid investment. Figure 1 illustrates this with the aid of a stylised yield curve.  In the example, a bank issues a loan at 3.5%, matched with bank deposits of shorter maturities offering an interest rate of 1%.  If the long rate rose to 5%, it would steepen the yield curve, increase the interest rate spread between lending and borrowing, and increase the NIMs.

It is worth noting that we wouldn’t expect this theoretical relationship, between the slope of the yield curve and NIMs, to hold perfectly in the real world.  For example, NIMs capture much more than just the gains of maturity transformation. For example, NIMs also reflect the rewards banks collect for bearing different types of risk (e.g. credit risk).

Stemming from this understanding of maturity and liquidity transformation Bill English  observes that this intuitive positive relationship has been the conventionalwisdom for some time.

We find no systematic positive relationship between the slope of the yield curve and NIMs

However, a very simple plot of the slope of the yield curve and the NIM does not deliver a positive relationship (Figure 2).  Instead, the slope goes the wrong way – it is negative for the UK confirmed by a simple regression – suggesting that an increase in the slope of the yield curve lowers the NIM. Indeed Table 1 (below) shows that this negative relationship arises in all countries in our sample bar the US, a point observed by a Liberty Street Economics blog post.

It is worth noting that in recent decades the countries in our sample have been through large economic, structural and policy changes, such as the introduction of inflation targeting, and changes in competition, financial liberalisation and regulation. These changes no doubt will have some impact on the slope of the yield curve and its relationship with NIMs, but those are beyond the scope of this article.

2) So how do interest rates affect NIMs?

Motivated by this discovery, we sought to inspect how the individual components of the slope of the yield curve (the short and long rate) affect NIMs. We find that the long rate is more important than the short rate in determining NIMs in a very simple regression model.  The long rate has a higher coefficient and is statistically significant for most countries.  The short rate is closer to zero and is insignificant for most countries, apart from Italy and Spain. Overall though, we find that a steepening of the yield curve is generally associated with a fall in the NIM (Table 1).

From this we conclude that, when it comes to interest rates, the long-term interest rate (unlike the short-term interest rate and the slope of yield curve) has a substantial positive impact on bank NIMs.  This finding helps to explain why an upwards parallel shift in the yield curve is good for net interest margins (because while the slope does not change the long rate goes up).  It is worth remembering that the results are driven by the average maturity and composition of assets and liabilities of bank balance sheets.  If the structure of their balance sheets changes, so too might these results.

What do these findings tell us about the past and the future?

The long rate in many economies has fallen gradually over time since the late 1980s.  Our simple empirical results suggest that there would be a corresponding fall in bank NIMs.  Figure 3 shows that while that relationship held in the UK prior to the financial crisis, it appears to have broken down since – as the NIM has flattened out in recent years, despite the continued fall in the long rate.  This isn’t just a UK phenomenon — NIMs in other countries have remained relatively stable since the global financial crisis too, despite falling long-term interest rates in these economies (Figure 4).  This may be because of the large macroeconomic and financial shocks that affected banks, or because banks have changed their business models and the structure of their balance sheets.  This is beyond the remit of this article. In light of this caveat it is hard to say with certainty whether this observed relationship between long rates and NIMs will reinstate itself or not; it is too early to tell.


Some central banks, such as the Fed and the Bank of England, have started the tightening phase of monetary policy, which has been associated with a steepening of the yield curve.  The commonly held view is that such a steepening of the yield curve should be unequivocally good news for bank profitability because it raises banks’ net interest margins.  This article challenges that conventional wisdom.  Using data for a panel of 10 countries over four decades, we find no systematic positive relationship between the slope of the yield curve and bank net interest margins.  Instead, we find that long-term interest rate tend to drive bank margins.  But even this latter relationship has weakened since the global financial crisis.  This suggests there is much uncertainty about the future relationship between interest rates and bank profitability.

Note: Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.

So You Want To Be An International Financial Centre?

From The Bank Underground.

The UK has a comparative advantage in financial services. But specialisation in this activity brings with it the challenge of the large gross capital flows that are linked to financial services exports.

The modern financial services industry allocates global capital flows through its balance sheets. Crudely speaking, profits correspond to a percentage over the value of flows, especially (volatile) banking flows, as banks arbitrage between assets and liabilities in different countries.

The chart captures this relationship by comparing the assets generated by banking flows (relative to GDP) – a measure of financial openness – with financial services exports (also relative to GDP). Countries which host international financial centres (the green dots in the chart), such as London for the UK (the red dot), are amongst the most open in the world.

Crucially, the chart is not capturing a mechanical effect. In the UK, for example, only one sixth of the statistical estimate for financial services exports is derived indirectly from international investment position statistics. The bulk of it is obtained from surveys conducted with banks. The estimate is also not affected by recent revisions to the UK national accounts.

International financial centres are compensated for providing essential financial services to the rest of the world. But the flipside is the need to absorb and manage the potential risks from volatile capital flows.

Note: Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.

The Problem With Productivity – Is The Finance Sector To Blame?

Silvana Tenreyro, External MPC Member, Bank of England, spoke on “The fall in productivity growth: causes and implications” as the 2018 Maurice Peston Lecture.

She explores the problem of low productivity since the GFC, and using UK data shows that the brake on productivity growth is from finance, manufacturing followed by ICT and services. But finance appears to be the number one sector causing the problem. It had the fastest-growing labour productivity of any sector in the run-up to the crisis, at 5% per year. Since 2009, productivity has actually shrunk by 2.1% per year. Indeed, key contributors to the crisis itself – risk illusion and increasing financial-sector leverage – may have increased (correctly measured) pre-crisis productivity growth.

Reading from this, as the finance sector continues to respond to pressure on margins, increased regulation, and lower growth, we think it will continue to be a brake on productivity, and it is possible that the growth of financial activities somehow crowded out the growth in the rest of the economy in a competition for talent and resources. Echoes of our recent discussion on Zombie firms! Relying more on the finance sector for growth looks like a problem.

Here is a summary of the speech.

Though commentators have referred to different measures of productivity, most have focused on aggregate labour productivity, defined as the total value added of the economy divided by the total number of hours worked.

Productivity matters for welfare. Over time and across countries, higher productivity is reliably associated with higher wages, higher consumption levels and improved health indicators.

Productivity is crucial to setting monetary policy. The MPC’s remit sets out a 2% inflation target over an appropriate time horizon with the rationale that inflation stability can lay the foundations for strong and sustainable growth. Productivity growth is the key determinant of how much demand can grow without creating inflation and hence it is a critical input into our forecast and deliberations.

The blue solid lines show a scenario where a 1% growth rate for potential productivity was overly pessimistic. The blue solid lines show a scenario where a 1% growth rate for potential productivity was overly pessimistic.

Over the three decades before the global financial crisis, productivity growth averaged 2.3% per year. Productivity fell in 2008 and 2009 as the financial crisis hit, and, in the seven years since, it has only grown by an average 0.4% per year. As a result, the typical worker in 2016, while still twice as productive as the 1970s, could only produce 1% more than in 2007.

Focusing just on the past half-century, the decade since the crisis looks like an aberration. Productivity growth barely deviated from its 2% trend until 2007 (Chart 2). It is little wonder, therefore – looking at these data – that forecasters (the Bank of England included) consistently predicted that productivity growth would recover to a rate close to its 1970s-2000s average.

Over a longer sweep of history, the past decade is far from unusual. Chart 3 shows annual UK labour productivity growth since 1760. Prior to the 1970s, there were often large shifts in the average growth rate of productivity from one decade to the next. Depending on how you interpret the chart, that could be a good-news or a bad-news story.

The ‘glass half full’ reading might note that we have been through several temporary periods of weak productivity growth before, but have always recovered. But there is also a ‘glass half empty’ interpretation. Robert Gordon from Northwestern University has argued that the hundred years spanning from 1870 to 1970 were exceptional in the number and scope of life-changing break-through innovations and there is absolutely no reason to expect growth to be as high and broad-based now. The progress since 1970, he argues, has been concentrated in a relatively narrow part of the economy: entertainment, communication and information processing. But in other essential areas like food, clothing and shelter, progress has been much slower.

Cross-country comparisons are tricky, but the ONS estimates that compared to the UK, labour productivity is on average 18% higher in the other six members of the G7, 28% higher in the US and 35% higher in Germany (Chart 4). These are significant differences. If British workers were able to catch-up to the G7 average, what currently takes us five days’ work to produce could be done in little over four. If we were able to catch up to Germany, we might all be able to go home from work on Thursday afternoon each week without any fall in GDP.

The plots have illustrated the UK productivity slowdown, both relative to other countries and also relative to the UK’s own recent past.

To attempt an answer, why productivity got lost, it is helpful to carry out a sectoral analysis, breaking down the productivity slowdown by industry. The sectoral distribution of productivity growth can help us locate where it has slowed. The slowdown, or difference in the aggregate productivity growth rates between the pre- and post-crisis periods for the UK economy amounted to (a negative) 1.5 percentage points. Remarkably, three-quarters of this productivity growth shortfall is accounted for by just two sectors: manufacturing and finance.

A further quarter of the slowdown is explained by two more sectors: information and communication technologies (ICT); and professional, scientific and technical services. The remaining 14 sectors contributed 0.5pp to productivity growth, both pre- and post-crisis. In other words, productivity outside those four sectors has been growing at a roughly constant, modest rate.

The finance sector is the biggest contributor to the productivity slowdown. It had the fastest-growing labour productivity of any sector in the run-up to the crisis, at 5% per year. Since 2009, productivity has actually shrunk by 2.1% per year.


It is unlikely that the entire slowdown in financial sector TFP is down to mismeasurement. A complementary explanation is that the key contributors to the crisis itself – risk illusion and increasing financial-sector leverage – may have increased (correctly measured) pre-crisis productivity growth. In doing so, they may also have sowed the seeds of the crisis and subsequent weakness. Increased leverage and higher risk tolerance boosted profits, earnings and output. That may have attracted capital and employees from other sectors of the economy. More broadly, rapid credit growth and low risk premia fed into higher asset prices, with positive spillovers to demand elsewhere in the economy. As the crisis hit, these channels went into reverse, leading to falls in wealth and higher uncertainty. Both lowered spending and output and probably also increased households’ labour supply.

Whatever the ultimate trigger of the finance-sector slowdown, its contributions to measured GDP and productivity growth are unlikely to pick up to those we saw in its pre-crisis boom. To achieve that would require a repeat of the type of unsustainable increases in leverage that we saw in the 2000s. The sector’s post-crisis performance has been as poor as its pre-crisis performance was strong. Credit and deposit growth have been weak as banks and households have sought to deleverage. It is possible that the growth of financial activities somehow crowded out the growth in the rest of the economy in a competition for talent and resources.

The financial-stability reforms we have seen since the crisis were put in place precisely to prevent the damaging consequences of those episodes.

Note: The views are not necessarily those of the Bank of England or the Monetary Policy Committee.

Who’s driving UK consumer credit growth?

From Bank Underground.

Consumer credit growth has raised concern in some quarters. This type of borrowing – which covers mainstream products such as credit cards, motor finance, personal loans and less mainstream ones such as rent-to-own agreements – has been growing at a rapid 10% a year. What’s been driving this credit growth, and how worried should policymakers be?

For many years regulators have relied on aggregated data from larger lenders to monitor which lenders and products are driving credit growth. These data are useful. But they also have important gaps. For example, they don’t include less-mainstream products that people with low incomes often rely on.

Crucially, such data do not show who is borrowing, or people’s overall debts across different lenders and products. This matters. If people borrow on many products, problems repaying one debt could rapidly spill over to others. Consumer surveys can offer some insights here. But surveys often have limited product coverage, are only available with a lag, and may suffer from misreporting.

To build a better, fuller picture of borrowing, the FCA requested credit reference agency (CRA) data for one in ten UK consumers. CRAs hold monthly data on most types of borrowing – including consumer credit, mortgages, and utilities. These data are really rich, going back six years, and can be studied at many different levels. For example, it is possible to scrutinise individual borrowing across products, or to focus on particular lenders or types of products.

We examined these data to assess possible risks from recent credit growth. This article summarises three particular insights which have emerged from this work:

  1. Credit growth has not been driven by subprime borrowers;
  2. People without mortgages have mainly driven credit growth;
  3. Consumers remain indebted for longer than product-level data implies.

Insight 1: Credit growth has not been driven by subprime borrowers

CRA data enables us to examine the distribution of credit scores among groups of borrowers. This is valuable because credit scores are excellent predictors of which types of borrowers are most likely to default or have high risks of suffering broader financial distress. A lower credit score indicates a greater risk of a person being unable to repay their debt. Those with very low credit scores are often referred to as ‘subprime’ borrowers.

In Figure 1 we show the share of outstanding consumer credit debt (net of repayments) by people’s credit scores. We divide the range of credit scores into ten buckets – the lowest bucket contains people with scores in the bottom tenth of the range (the riskiest borrowers).

Doing so reveals that a small proportion of all consumer credit debt is held by subprime consumers. There are some important differences when we compare people holding different credit products. Borrowing on credit cards with 0% offers and motor finance is concentrated among people with the highest scores. This contrasts with people borrowing on interest-bearing (non-0%) credit cards who more commonly have low scores.

Given motor finance and 0% credit cards have accounted for a majority of consumer credit growth since 2012, this suggests much of the growth is going to the borrowers least likely to suffer financial distress. This story is consistent with high-cost credit markets used by subprime borrowers not rapidly expanding – on the contrary, some are contracting.

In Figure 2, we turn to how the distribution of borrowing has changed over time. Here we find little difference in credit scores over the recent period of rapid credit growth. This holds when looking at both the outstanding stock and the flow of new borrowing. At face value, this indicates that lenders have not dramatically relaxed their lending standards. But observing a similar credit score distribution when the macroeconomic environment has slightly improved may be better interpreted as a deterioration. The only product where we find an increased concentration of subprime borrowing is interest-bearing credit cards.

History also offers some caution on the relative importance of subprime lending. Recent research on the US mortgage crisis found the pre-crisis growth of subprime borrowing was less dramatic and important to explaining the crisis than earlier studies implied.

Insight 2: People without mortgages have mainly driven credit growth

The recent credit growth has followed a tightening of mortgage lending requirements. Did this tightening have the unintended side-effect of turning mortgage borrowers away from extracting home equity and instead towards consumer credit?

We assess the interaction between these two markets by splitting the growth and stock in borrowing between mortgagors and non-mortgagors. This is shown in Figure 3. About half of outstanding consumer credit is held by those with mortgages. However, this group accounts for a minority of growth in credit balances, with 60% of the growth in credit balances coming from non-mortgagors.

It is comforting that mortgagors do not appear to be bypassing tighter mortgage regulation by amassing consumer credit debt. But a key question going forward is how much of the growth is coming from renters and how much from outright owners.

We know that renters tend to spend a higher share of their income on housing than mortgagors, and so may have less income available for debt repayments. Rapid increases in indebtedness among renters could therefore be a vulnerability.

It is also possible that outright owners are taking out credit, even if they don’t need it. Survey data suggest around 40% of households with consumer credit debt hold savings in excess of such debt. If driven by outright owners, rapid credit growth among non-mortgagors may be less worrying.

Insight 3: Consumers remain indebted for longer than product-level data implies

The Bank has previously argued that lenders’ consumer credit portfolios turn over relatively quickly, reflecting the short terms of consumer credit products (relative to mortgages). In theory, this rapid turnover means that the prudential risks from outstanding consumer credit could quickly increase (or decrease) if lending standards were to deteriorate (or improve).

While this may hold from a lender perspective, our analysis tells a different story from consumers’ perspective. We find that although a consumer may clear their debt on one credit product, it is not uncommon for them to remain in debt as they transfer balances, take out new credit products or draw down on existing credit lines (such as credit cards). As shown in Figure 4, 89% of the total outstanding stock of debt in November 2016 was held by people who also owed debt two years earlier. While approximately half of new borrowing is due to ‘new’ borrowers, these people are typically only able to access relatively small amounts of credit and therefore account for a small proportion of the overall stock of debt.

An implication of these findings is that regulators should not become too relaxed when they observe improvements in specific products at particular lenders. Unless the borrower population significantly changes, it is possible that the risk of consumer harm will simply be shifting from one part of the market to another rather than reducing. It is therefore important to regularly examine the financial health of people and their debts holistically using CRA (or similar) data.

Should policymakers be worried?

Credit growth not being disproportionately driven by subprime borrowers is reassuring. As is the lack of evidence that mortgage lending restrictions are pushing mortgagors towards taking on consumer credit.

But vulnerabilities remain. Consumers remain indebted for longer than previously thought. And renters with squeezed finances may be an increasingly important (and vulnerable) driver of growth in consumer credit.

Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.

How Does Uncertainty Affect How UK Firms Invest?

From the Bank Underground Blog.

Uncertainty is in the spotlight again. And the MPC believe it is an important factor influencing the slowdown in domestic demand (August 2017 Inflation Report). Previous work by Haddow et al. (2013) has found a composite aggregate indicator of uncertainty combining several different variables that does appear to have explanatory power for GDP growth; but as Kristin Forbes notes these measures correlate better with consumption than investment. So in this blog post, we look at firm-level data to explore measures of uncertainty that matter for how firms invest in the United Kingdom. Our aggregate measure of uncertainty has a better forecast performance for investment than the composite aggregate indicator does.

Uncertainty is difficult to define, and since it is unobservable, it is hard to measure. It is correlated with other factors that drive an economy but it is unclear whether movements in uncertainty simply reflect the influence of those factors or indicate an independent change in uncertainty (see, for example, Bloom (2014)). This makes it challenging to distinguish its direct effect on the economy. Various approaches have been adopted in recent literature both to measure it and to capture its effects on macroeconomic variables, like GDP, consumption and investment. Jurado et al. (2015) have emphasised certain desirable features of uncertainty measures. First, these measures should be forward-looking since we are interested in uncertain events in the future. Second, they should measure variation in future outcomes (i.e., the width of the probability distribution) rather than specific future outcomes (like a mean or a median). And third, they should not include a component that can be systematically forecast in advance; we are not typically uncertain about outcomes that we can forecast!

One potentially useful measure of uncertainty is based on UK firms’ stock price volatility, inspired by a similar measure introduced for US firms by Gilchrist et al. (2014). The intuition behind the stock market uncertainty (SVU) measure is that we would expect those firms facing a more uncertain environment to have a more volatile stock price. This is because investors are less certain about the value of the company and hence the value of the shares they hold in the company. We use this measure to construct an estimate of UK firms’ uncertainty that has all three of the desirable features mentioned above.

To calculate the SVU measure we use a sample of 622 firms listed on the stock exchange in the UK. To do this, we first estimate the daily variation in individual stock prices that cannot be explained by general market variation in a capital asset pricing model (CAPM) (see, for example, Sharpe (1964)). Then, we calculate the quarterly firm-specific standard deviation of the daily unexplained returns from the first step. The outcome is a firm-specific and time-varying SVU measure. Finally, we estimate a common component of these measures over time, which we interpret as an aggregate-level measure of the SVU.

Because the SVU measure already reflects the expected variation in the stock price based on general market variation in the first step, this measure picks up the unforecastable variation in the price. That helps achieve the desirable features of uncertainty measures discussed above; it is forward-looking (stock prices should include information about future prospects of a firm), it measures variation rather than means or medians, and it makes an attempt to exclude the forecastable component of uncertainty.

Chart 1 shows what the resulting aggregate SVU measure looks like. It also shows how it compares across large firms (over 250 employees) and small and medium-size firms (about 20% of firms in our sample).

The aggregate uncertainty measure is fairly volatile, and there is a large spike during the financial crisis. The SVU measure for small and medium-size firms is even more volatile than for larger firms, probably reflecting the more uncertain environment facing smaller firms. When compared to the Haddow et al. uncertainty indicator, the measures generally move in the same direction. More recently, both increased in anticipation of the EU Referendum, but after it, the SVU measure fell while the Haddow et al. indicator remained elevated (Chart 1).

Chart 1: Different measures of uncertainty

So, what might be the effect of uncertainty on investment? Chart 2 shows that firms which experienced higher uncertainty during the crisis, on average, have subsequently tended to report lower investment (in other words, the distribution for these firms is, on average, more to the left than for the other group of firms). This is in contrast to firms which experienced less uncertainty during the crisis, on average, and have tended to invest more since the crisis. The difference between two groups of firms is statistically significant and suggests that investment dynamics might be negatively affected by uncertainty.

Chart 2: Distributions of firm-level investment-to-capital ratios split by increases in uncertainty during the financial crisis

The chart shows sample distributions for investment-capital ratios since 2009 by the size of an increase in uncertainty during the financial crisis.

But Chart 2 is only suggestive. It does not prove any form of causality. To get a better idea of how uncertainty might actually affect investment, we model investment at the level of the firm. We also control for other relevant drivers of investment, as well as variation in investment across time and across firms. More specifically, we regress firm-level investment-to-capital ratios on variables like firm-specific sales, a proxy for future investment opportunities, and different measures of the cost of capital. We find that firm-specific SVU is a very significant explanatory variable for firms’ investment rates in all specifications of the model that we tried, as well as when we include different measures of the cost capital and firm-specific risk premia. We also find that the SVU measure has only been important, in a statistical sense, after the financial crisis.

Given there is evidence that firms’ investment behaviour is correlated with our measure of uncertainty, we can use that insight to study aggregate business investment in the UK. We model uncertainty in a multivariable time series model that also includes other relevant aggregate level variables (GDP, interest rates and inflation). In this framework, an increase in uncertainty produces a relatively persistent effect on investment (Chart 3). This effect peaks after one year and then gradually dies out over the next two years. And a model with our SVU measure forecasts business investment dynamics better over the past than a model with the Haddow et al. uncertainty measure (Chart 4).

Overall, there is evidence that our measure of firm-specific uncertainty can help explain investment behaviour both at the level of the firm and in aggregate. Whereas the work by Haddow et al. suggested that the uncertainty is an important driver of GDP fluctuations, our work provides complementary analysis, using a different measure of uncertainty, to suggest that uncertainty is a crucial factor in firms’ investment decisions.

Chart 3: Impulse response of business investment to an uncertainty shock

The chart shows percentage changes to a one-standard deviation shock to the uncertainty variable. The impulse response has been identified with a Choleski ordering of the variables (see here for more detailed definitions), where investment reacts with a lag to exogenous shocks in the other variables.

Chart 4: Relative forecast performance of different uncertainty measures on business investment

The chart shows the root-mean square error (RMSE) relative to a random walk forecast for annual changes in business investment at different forecast horizons, where random walk = 1. The lower the RMSE, the better the forecast.

Marko Melolinna works in the Bank’s Structural Economic Analysis Division and Srdan Tatomir works in the Bank’s Macro Financial Analysis Division.

Note: Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.