In my last column (Forecasting inflation accurately, FE, July 24), I looked at how well experts do in forecasting India?s inflation rate. RBI?s surveys of experts are relatively recent, so there was not much data to go on. The exercise was motivated by the surge in inflation and the apparent difficulty in predicting future price level rises. Ideally, one would have a detailed model of the economy that uses all kinds of information about markets, aggregate supply and demand to get the best forecasts. Perhaps, the experts have their own models that they use to make the forecasts reported to and by RBI.
Complex models require a lot of information that may not be available in time for making forecasts. It is interesting to ask if one can do reasonably well using just very simple information. For example, this period?s inflation might tell us what inflation will be in the next period. Another possibility is that what is happening in bond markets can give us some useful information.
Bonds provide returns that are fixed in nominal amounts. A bond that offers a rate of 10% will provide income of Rs 10 annually for every Rs 100 invested, until the bond matures and the principal is paid. If market interest rates fluctuate, the price of the bond may vary, so that the income provides a return equal to the market rate or bond yield. Investors will also care about risks of not being paid, but this is not a problem for Indian government bonds, since there has never been a default. But inflation can erode the value of the income generated by a bond, so when inflation is expected to be high, bondholders may want a higher market interest rate to stay invested. Bond yields may provide information on expected future inflation. For example, 10-year Treasury bond yields in the US are only about 3%, reflecting beliefs that inflation will stay low. If investors expect inflation to go up, they will demand higher yields to maintain the same real rate of return. If inflation expectations are right on the whole, then they will be good forecasts of future inflation.
The calculations are complicated by all the other factors that can affect markets for bonds, including the expected performance of other financial assets and beliefs about the future course of the real economy, which will affect the rewards that various financial assets provide. This brings us back to a complex model, which we may want to avoid. In the US, forecasters are able to extract information about inflation expectations from the overall pattern of returns on government securities of different maturities (the yield curve), as well as the relative returns on regular Treasury bonds and corresponding inflation-indexed bonds. The difference in those two returns should provide a good indicator of inflation expectations or the market?s forecast of future inflation. The ?market?, of course, is an aggregator of various participants? beliefs, whether they are experts or not.
India does not have well-developed bond markets and does not offer inflation-indexed bonds. So, one cannot perform a sophisticated analysis of what relevant information bond market yields might provide for predicting inflation. Still, it is worth looking at the data we have easily available.
I took 10 years? data covering 1999Q4 through 2009Q3 and calculated average (WPI) inflation and 10-year government bond yields for each quarter. Ideally, one should go to higher frequency data but the exercise is illustrative. I first tried to see if bond yields could tell us anything about future inflation rates. They did not, whether looking at one, two, three, four or eight quarters ahead. On the other hand, the inflation rate itself captured half the variance of the next quarter?s inflation rate. Adding the bond yield did not increase the predictive power of the model. But adding the change in the bond yield did make a small but significant difference. Increasing bond yields, according to this simple model, would be an indicator of higher inflation a quarter ahead.
Unfortunately, when one includes the inflation rate from two quarter?s past as well as the last quarter, the additional predictive power of the change in bond yields goes away. So, this simple exercise is ultimately inconclusive. It could be that bond markets in India are underdeveloped or distorted in ways that make their behaviour of little or no use for forecasting inflation. Still, the idea that markets aggregate information relatively efficiently is appealing (recent bubbles and crashes notwithstanding) and it would be nice to see a more rigorous attempt to combine inflation history, market information and expert judgements in making inflation forecasts. And of course, deeper, more efficient bond markets are desirable irrespective of their contribution to forecasting inflation.
The author is professor of economics, University of California, Santa Cruz