Friday, December 10, 2010

What are the Costs of Inflation, and How Should We Measure It?

As inflation and its costs could become more important to us soon, I thought it would be useful to consolidate some key ideas. As well, this is much more interesting to me than whether people in Congress want to, or do not want to, tax rich people.

During the 1970s, prior to the Volcker disinflation of the early 1980s, economists and the public-at-large seemed to be having a hard time understanding the costs of inflation. There was public dissatisfaction with high inflation rate at the time, but the reasons for this dissatisfaction seemed vague, and had little to do with what was known about the economic costs of inflation. During this period, someone told me to read this paper by Robert Solow. Solow came to the conclusion that the moderately-high inflation rate in the US during the 1970s was not such a big deal, and that the best strategy was to just live with it. Obviously Solow's views did not prevail, and the Volcker disinflation was much less costly, in terms of lost economic activity, than Solow and others anticipated.

Since the Volcker disinflation, inflation in the United States has been low-profile in the minds of most people, outside of policymakers in the the Federal Reserve System and some groups of monetary economists. However, given our recent central banking experiments, which have greatly expanded the Fed's balance sheet, and increased the probability of higher inflation, everyone should have inflation on their minds.

The costs associated with inflation are as follows:

Anticipated Inflation: This is by now well-understood. In a host of monetary models, inflation causes price distortions that misallocate resources. These distortions are associated with the use of non-interest-bearing money as a medium of exchange. The higher the rate of inflation, the lower the real rate of return on money, and the less money we want to hold, in real terms. We then have less real money balances to support exchange, leading to lower consumption and lower aggregate output. The costs of anticipated inflation are straightforward to correct with monetary policy. Milton Friedman told us in his 1969 "Optimum Quantity of Money" essay how to do this. Essentially, the idea is to conduct monetary policy in such a way that the nominal interest rate is always zero, in that a positive nominal interest rate typically represents the presence of a distortion. This is a "Friedman rule," whereby a zero nominal interest rate acts to equate the rate of return on money with the rate of return on other safe assets, and this can be achieved with trend deflation.

There are different ways to implement the Friedman rule, though. For example, an alternative to trend deflation is to pay interest on money at some market rate of interest (say the interest rate on Treasury bills). Of course, the stock of money issued by the Fed consists of two components, currency and reserves, and it does not seem feasible to pay interest on currency. However, interest can be paid on reserves, as has been the case in the United States since late 2008. Indeed, part of what motivated the Fed to convince Congress to permit the payment of interest on reserves was Friedman-rule considerations. Note, though, that the reserves on which the Fed, and other central banks, pay interest are only reserves held overnight. Financial institutions do not earn interest on their daylight reserves, though in the US they pay interest on negative balances, i.e. "daylight overdrafts." In principle, daylight reserves should also bear interest. For example, if financial trading took place on a 24-hour basis so that there was no overnight shutdown in trading (as may well be the case in the not-too-distant future), it seems clear that payment of interest on reserves should take place continuously.

Sticky price distortions: New Keynesians focus, sometimes exclusively, on costs of inflation related to price stickiness. If there is general price inflation, if firms change prices infrequently, and if firms do not coordinate price changes, then there will be relative price distortions. Firms selling the same goods and services under the same conditions will in general be charging different prices, and this is economically inefficient. There is also a monetary policy fix for this problem, though it is different from the Friedman rule prescription for eliminating the positive-nominal-interest-rate distortion. Typically, price stability - achieved with the appropriate monetary policy rule - acts to eliminate sticky-price distortions. If the average level of prices is not changing over time, then firms need only change their prices due to real factors affecting their production technologies, the costs of inputs, or the relative prices of outputs, and this is efficient.

Unanticipated Inflation: Older discussions of the costs of inflation, for example in the Solow piece discussed above, tended to view the effects of unanticipated inflation only as unanticipated redistribution with no particular efficiency implications. The idea is that, since most debt contracts are written in nominal terms, an unanticipated inflation will tend to redistribute wealth from lenders to borrowers. In terms of aggregate welfare, this should not be something we would get too excited about. However, I think economists are beginning to take the costs of unanticipated inflation more seriously.

Generally, there are two problems with inflation uncertainty. One is an ex post problem, and the other is an ex ante problem. First, the ex post costs result from the fact that default is costly. For example, an inflation rate that is lower than expected will increase the real debt burdens of borrowers, increasing the likelihood of default. Default is costly, as it can lead to legal costs from bankruptcy, or to the transfer of collateral, which tends to be an inefficient reallocation of assets (because the borrower values the collateral more than the lender does). It can also be costly for the inflation rate to be higher than expected. This reduces the real value of debt, which could ultimately lead to default as well. Why? Firms and financial institutions can simultaneously be lenders and borrowers. If the value of what is owed to me falls in real terms, this may impair my ability to pay off my creditors, possibly causing me to default on my debts, further impairing the ability of others to make good on their debts. As well, consider a case where the inflation rate rises in an unanticipated manner, and then stays at this higher level for a long period of time. Through a Fisher effect, this would tend to increase all nominal interest rates. This tends to squeeze financial intermediaries, for example banks, that borrow short and lend long, and these financial intermediaries could end up defaulting.

In terms of ex ante costs, inflation uncertainty will tend to reduce credit market activity, since both borrowers and lenders will want to reduce their exposure to unanticipated changes in their net wealth due to unanticipated inflation or deflation.

Benefits from Anticipated Inflation: The key costs of anticipated inflation fall on the holders of central bank liabilities - currency and reserves. But in some cases, we may not care so much that these liability-holders are suffering. A large fraction of US currency is held either outside the United States, or by people who are using it for nefarious purposes. Inflation is a tax on central bank liabilities, and perhaps the taxation of people who are not US residents or who are trading illegal drugs or evading income taxes would be a good idea.

Now, no one has written down a quantitative model with all the features I have described above, which can tell us the optimal inflation rate that the central bank should deliver. The Fed seems to have converged on 2% as an implicit inflation target, but that seems based on experience more than anything, i.e. with a 2% inflation rate, no one seems to be complaining much.

Inflation Measurement
Now, there are many ways to measure the inflation rate. Different approaches to measurement can produce quite different inflation numbers, so it matters practically which inflation measure the central bank chooses to control. What do the considerations above tell us about this choice?

In New Keynesian sticky-price models, if some prices are flexible, while others are sticky, then the only prices we care about are the sticky prices, since it is the price stickiness that is leading to inefficiencies in such a model. It is tempting, then, to argue that the price index we should be interested in should be one that excludes goods and services with volatile prices. The presumption is that sticky-price goods will have prices that are less variable than the prices of flexible-price goods. In principle, this need not be the case, since the price of a good could change infrequently, and yet the variance of that price could be relatively high, due to the fact that price changes are very large when they occur. In practice, however, the consumer price index excluding food and energy (the "core" CPI) seems to include what a New Keynesian would consider sticky prices, while excluding the flexible ones.

However, what about the costs of inflation that have nothing to do with price stickiness? Why would those other considerations cause me to eliminate some prices from consideration in my price index? Maybe there are other prices I want to include, other than what is in the CPI, or in some other measure of the price level?

Consider the costs of anticipated inflation. Central bank liabilities are being used not only to purchase final goods and services, but to settle debts, to purchase assets, and to purchase intermediate goods and services. Thus, how individuals, firms, and financial institutions choose to economize on their holdings of central bank liabilities will depend not only on the prices of final goods and services but on asset prices and the prices of intermediate goods and services. How should we take that into account? The key issue is that many assets, including money, may be bought and sold in a chain of transactions by which a consumer reallocates consumption intertemporally. You see this, for example, in some models of payments arrangements, for example the payments models that Scott Freeman worked with in the 1990s. These considerations introduce other kinds of distortions which are determined in part by how economic agents participate in financial markets, and by central bank intervention, for example in terms of daylight overdrafts. Some of these considerations could lead us to include more prices in our measure of inflation, rather than fewer.

How would the costs of unanticipated inflation be reflected in the choice of a price index for a central bank to focus on? Here, the issues are closely related to those we discussed in connection with anticipated inflation. The costs that arise have nothing to do with whether individual prices are sticky or flexible, smooth or volatile. There may be good reasons to include intermediate goods prices, or prices of final goods and services that are not consumption goods. Further, unanticipated-costs-of-inflation considerations should lead central banks to prefer price level targeting over inflation rate targeting. With pure inflation rate targeting, history does not matter so that, for example, the Fed targets this year's inflation rate at 2% whether last year's inflation rate was 0%, 5%, or 3%. Price level targeting requires that the central bank adjust for previous errors. If the actual inflation rate for the current year falls below the 2%, the central bank targets next year's inflation rate at higher than 2%. In this fashion, we know what inflation rate to expect over any horizon, and debt holders can be more certain of realized rates of interest over any maturity.

Where does this leave us? We cannot determine the appropriate inflation measure for a central bank without constructing a model that captures all of the key factors above, and without knowing what the central bank's objectives are. However, I think we can at least say that exclusive focus on measures of "core" inflation - the core CPI or core PCE index - seems wrongheaded. These measures were developed in the 1970s, seemingly to take the heat off policymakers by focusing on inflation measures that would yield lower inflation rate numbers. Before recent New-Keynesian justifications for core-inflation targeting, the typical argument was that movements in volatile prices, like those for food and energy, are more likely to be temporary, and we want to restrict our attention to price movements that are more likely to be persistent. Of course, some changes in prices that have been volatile in the past can be highly persistent, and some changes in in prices that have been very smooth can be temporary. Presumably it is preferable to make these judgements based on the economics rather than building this into our inflation measures.

I think it is easy to make the case for the use of the broadest possible measures of inflation. While traditional thinking might tell us that we should focus only on the prices of consumer goods, the fact that the prices of many other goods, services, and assets, can matter for the costs of inflation, through the chain of transactions, should lead us to consider broader measures. Even the Gross Domestic Product Deflator - essentially the price of final output in the United States - may not include all that we want in a price index.


  1. Perhaps you implied this, my experience in Latin America is that unanticipated inflation has two negative effects:

    1) it raises the real term premium to a level that chronically discourages l.t. investment.

    2) it raises the standard deviation of profit forecasts, again raising the equity risk premium and discouraging l.t. investment.

    The second effect happens because profit margins are quite difficult to forecast when input costs, interest rates and sales prices experience high volatility.

    The ultimate cost of high and variable inflation is low investment -- a "hollowing out" of the economy's productive capacity -- and chronically low growth. For instance, in the mid 90's, I toured Brazilian steel plants that had been starved of maintenance capital for years, rusting hulks that, without substantial capex, could only operate at below 50% of capacity.

  2. Professor,

    In most NK models, is it first assumed that prices are sticky and then the rest of the model is formulated? Or, are sticky prices derived as a feature of the model?

  3. There is state-dependent pricing. You assume that there is a fixed cost of changing a price, and a firm does it when it is optimal. There is also time-dependent pricing (e.g. Calvo), where a firm changes its price at some frequency or when it gets a random opportunity to do it. So, it depends what you mean by "modeling" the sticky prices. With state-dependent pricing, you don't show where the cost comes from, and with time-dependent pricing, there is also a lot that is left unexplained. The most common NK approach is to assume Calvo pricing, and go from there. That doesn't do it for me, but plenty of people disagree with me.

  4. This comment has been removed by the author.

  5. "Firms and financial institutions can simultaneously be lenders and borrowers. If the value of what is owed to me falls in real terms, this may impair my ability to pay off my creditors, possibly causing me to default on my debts, further impairing the ability of others to make good on their debts."

    I don't understand this. If the real value of what's owed to me has fallen because of a general rise in prices, then the real value of what I owe must have fallen also. The only situation in which this would matter is if one creditor's assets are particularly tied to some (sticky) relative price, which becomes distorted as a result of the inflation.