Prudential Regulation: Are Government Bonds Safe

This article is about why do we regulate banks and why do we have prudential policy. Moreover it’s about what happens when prudential regulation says that government bonds are safe and in reality they are not.

This is a storytelling paper. Perhaps, some of you will see this as a horror story. I prefer to look at it as a cautionary tale.

This story goes back to the basics: Why do we regulate banks? Why do we have prudential policy?

The first reason is that banks play with other people’s money, and playing with other people’s money gives one an incentive to take on excessive risks.

Therefore, the regulation ensures that banks have enough of “their own skin” in the game by forcing them to have some capital.

Under a standard regulation policy, we typically identify  a certain security as a risk-free, and when agents invest in this safe security, they don’t have to put their skin in the game – this is the whole idea of risk coefficients. A typical safe security – the one which you don’t have to backup with your own capital – is a government bond.

However, we’ve come to realize that government bonds may not be as safe as they are considered to be by the prudential regulation with an example of some countries (Russia in 1998 and  Argentina in 2001 are clear examples, but this is not an exclusively emerging market phenomenon). So what happens when prudential regulation says that government bonds are safe and in reality they are not?

In this case, prudential regulation fails. It no longer prevents the banks from taking on excessive risks. And thus the banks would take on excessive risks in a very specific fashion. They would take on government bonds and things which are correlated with those.

The model in our paper captures this story and offers two key insights. First, if the regulator is at the same time the borrower, it has an incentive not to adjust the regulation, i.e., to classify its bonds as risk-free, – to keep the interest rates on them low. The reason for the lower interest rates is the willingness of the gambling banks to buy these bonds at less than actuarially fair interest rate. Hence, we highlight a novel moral hazard problem – not just that of banking, but that of banking regulation. And this is the main message of the paper.

The second key insight, perhaps more relevant to Europe, is that the cost of this moral hazard problem is not just the risk of financial crises. The other cost of the regulation failure is diversion of investment from productive users, from good projects, into less productive (risky) projects which are correlated with government bonds.

To make the model as simple as possible, let’s assume there are only three things in which banks can invest. These things are the good, the bad and the government. The good projects are those where investments should be going, they provide a safe return r. The gambles are high risk projects (they are like lotteries, they provide return R with probability p and zero otherwise, so their expected return is lower than that of “good” projects: pR < r < R), and the government bonds are assumed to be zero-risk and pay (1+ig). Note that investors themselves would never invest into “bad” projects but banks might because they have limited liability (and they invest other people’s money). Because of that, without a regulation, the banking system would invest only into “bad” projects (gambles).

In this model, nobody but the bankers can tell apart a risky project from a safe project. As a result all of the investment is going through the banks. The banks are subject to limited liability and moral hazard. We make the supply of deposits and the outstanding government debt exogenous.

It is clear that in the first-best allocation all of the investment would go into safe “good” projects (except for the resources going into government bonds, of course). If this were an equilibrium, the competition in the banking sector would lead the interest rates in the economy to be equal to the return on the safe project: r = 1 + i = 1+ ig.

But absent regulation, an individual banker in this model has an incentive to invest into a risky project to earn R-r (with probability p) since

p(R−(1+i)) > r − (1+i) = 0.

So, absent regulation, this model does not deliver the good (socially desirable) banking allocation. How do we fix this? We introduce capital requirements for the banks, i.e., we force them to have some of their own wealth invested into projects together with deposits.

In the model, the regulation has just two parameters: b – the fraction of the investment that has to be owned by a banker, and q – the fraction of total portfolio that has to be placed into government bonds. And as long as the loss from gambling exceeds the gain from gambling, i.e.

br(1−p)>(1−q)p(R−r),

prudential regulation does restore the first best – all of the resources flow into safe projects and government bonds. So things work as longs government bonds are safe.

Let’s see what happens when government bonds are no longer safe. Assume the probability of government default is exogenous and the same as the probability of a gamble not repaying (equal to 1-p). So government bond is going to look like just one of those gambles. But the regulation doesn’t recognize this fact, the regulation still treats government bonds as safe. When government bonds become just another gamble, prudential regulation fails because it’s now possible to satisfy prudential regulation and run a gamble. Gambling banks simply build a portfolio of government bonds and risky projects perfectly correlated with the government repayment.  And since government bond is just another gamble from these banks’ perspective,  it is going to pay R in equilibrium. Or rather, it promises to pay R, but makes that payment only with probability p.

Furthermore, banks that don’t gamble are driven out of the market because they cannot offer the same high interest rates as gambling banks or the government and thus cannot attract deposits.

1+ig=R > r

In this model, if prudential regulation is not adjusted, the equilibrium is very drastic – all household deposits are funneled into gambling banks, which put them into government bonds and risky projects which are perfectly correlated with government bonds, and the whole system crashes down when the government defaults.

If the regulator recognizes the riskiness of government bonds, even with risk neutral investors you would still have higher interest rate ( rp ) on the government bond than if you allow the gamble. Basically allowing banks to gamble makes it cheaper for the government to borrow. So, who is the real gambler here? The government. This is the message number one. Message number two – no household funds flow into good projects.

We next enrich the model to make it more realistic along two dimensions. First, we change the environment so that, even when regulations fails, not all banks are gamblers (we do so by endogenizing returns on safe projects). Second, and more important, alteration is making the regulation more realistic. We now explicitly make capital requirement b dependent on the share of bank’s portfolio invested in government bonds:

b(q)=(1-q) p(R-r)r(1-p).

The point of this exercise is to see if the key findings are robust.

What happens in the augmented model is something very dramatic. With the more realistic banking regulation, the (risky) government bond becomes much better than just a gamble – it becomes a permit to take on a deposit. So, the risky government bonds become even more attractive to gambling banks in this economy. And as a result, the interest rate that the government has to pay (when the regulation ignores riskiness of the sovereign bonds) fall even further. In fact, all interests rates in this new economy are higher than in the benchmark, except the interest rate on government debt, which is actually lower. Thus, the key finding that government has strong incentive to allow banks to gamble because this makes borrowing cheaper is not just confirmed, but further reinforced.

Now to the real life examples.

The Russian government in 1998 might not be the most experienced or sophisticated government, maybe they just forgot to adjust prudential regulation. But as for Argentina in 2001, they didn’t forget. In 2001, they changed the pension funds law to allow state pension fund to buy government bonds which were considered safe. So, Argentina is a clear case of the government explicitly changing the rules to allow the gambling.

The Russian example gives us a great anecdote and a way to check the story at the micro level. When I looked at Russian data, I wondered, what was correlated with government bonds at that point? Russian budget depended heavily on oil and gas exports, so I expected that gambling banks would invest heavily in these resource sectors, as those investments would be correlated with the government bonds. But Russian banks were much smarter than me, and they found something perfectly correlated with the government default – ruble. So what did Russian banks get into? They got into currency forwards, which are derivatives on currency futures. Russian banks were promising to pay you money if ruble collapses. How much were they promising to pay you? In case of «Inkombank» – around 719% of their own capital. And these banks tended to be the ones holding government bonds, thus satisfying prudential regulation.

We look at the micro level – at the correlation between exposure to currency risk and government bonds. We find that before the crisis, in the first two quarters of 1998, we had a positive correlation between these variables, which completely vanishes after the crisis. This correlation was much stronger for private and foreign-owned banks than for state banks, because presumably government has much stronger sway with state banks.

What are the implications for Europe from this model?

The LTRO (long-term refinancing operations) scheme introduced by ECB in 2011 allowed Eurozone banks to borrow from the ECB while pledging their home countries’ government bonds as collateral. And yes, these bonds were considered safe. This, of course, created an incentive for, say, Italian banks to buy Italian government bonds. It may be politically necessary for ECB to support buying south European countries government bonds. But this paper points to one additional cost of doing so via the banking system – it diverts investments from productive projects into less productive gambles correlated with government bonds.

The main conclusion from this story is that there should be a reasonable distance between the banking regulator and the sovereign borrower – so that the regulator does not have an incentive to allow the gamble.