Why Food Price Volatility Matters
In a recent issue of Foreign Affairs , Christopher Barrett and Marc Bellemare write a stimulating note on the recent price movements in agricultural commodity markets. They appear to have three clear messages for policymakers:
First, price levels and price volatility are different concepts. In recent months, markets have been characterized by very high price levels, but volatility “although high these past few years, is not out of line with historical experience and is generally lower than it was in the 1970s.”
Second, high price volatility does not hurt consumers since they can adjust consumption in response to different relative prices. It is high prices that have the most nefarious consequences, as they reduce consumers' purchasing power and can increase poverty and, in some cases, hunger and malnutrition. Hence, policymakers should favor and adopt policies that contribute to the reduction of price levels.
Third, high price volatility does “hurt” producers by reducing profits. However, if policymakers want to consumer welfare and reduce “political unrest,” it is price levels that they ought to worry about, not price volatility.
The authors argue further (based on a simple regression model) that there is a positive correlation between price levels and “political unrest” and, surprisingly, a negative correlation between volatility and “political unrest.” This corroborates the argument that it is price levels, not volatility, that hurts consumers.
Regarding the first message:
We praise the authors for explicitly stating what statisticians and economists, but not all policymakers, know well: price levels and price volatility are different concepts. However, we remind the authors that it is necessary in academic and policy discussions to agree on and use an adequate measure of price volatility. In fact, we have serious reservations about the authors' choice of a simple rolling average of 3 or 6 month unconditional standard deviations associated with FAO's monthly Food Price Index, the most important being that price volatility depends on past price levels in a fairly intricate manner. Put differently, an accurate measurement of price volatility depends on the construction of a statistical model that describes the evolution of volatility through time as a function of the past history of prices (see The Food Security Portal Excessive Food Price Variability Early Warning System ). In fact, past volatility may even have an impact on current volatility. A large and growing literature in empirical finance has taught us that ignoring a dynamic model of conditional volatility in favor of simple unconditional standard errors may lead to severe errors in measurement.
However, even if we use the authors' 3 or 6 month rolling windows to calculate unconditional price volatility for soybeans, wheat, and corn using daily future price contracts (shortest maturity), we observe significant increases in price volatility that are certainly not in line with the historical time series (our data extends as far back as 1959). The case for increased price volatility in recent years is particularly strong for wheat and corn. Thus, contrary to what the authors claim, we are unconvinced that current volatility is not high compared to that of the past.
Regarding the second and third messages:
We certainly agree that the impact of increasing agricultural prices on consumers' welfare may be substantial and of “first order” if compared to the impact of increased volatility. However, there are three important caveats that need to be taken into consideration. First, increased price volatility is positively correlated with producers' expected losses. As such, high price volatility may reduce and distort input allocation into agricultural production, reducing the "supply curve" and increasing price levels. This, as forcibly argued by the authors, hurts consumers. Hence, identifying periods of increased price variability in a transparent and consistent manner may prove very useful for policymakers, even if their primary concern is only the protection of the most vulnerable consumers.
Second, most of the arguments advanced by the authors regarding the impact of high price levels on consumers apply mostly for urban areas, but not for rural areas where households might be both consumers and producers of agricultural commodities. In this case, increased price volatility might increase expected losses, directly impacting households’ consumption decisions. Finally, increased price volatility through time generates the possibility of larger net returns, and potential larger returns create the possibility of constructing investment portfolios that previously did not contain agricultural commodities. Thus, increased price volatility may lead to increased (potentially speculative) trading (for example, new indices including food commodities have increased from USD 13 billion to USD 260 billion between the end of 2003 and March 2008).
We do agree with the authors that most of the proposed steps to reduce price levels may help reduce price variability. However, specific steps can be taken to help producers better protect themselves against price risk and consequently encourage production. In the absence of markets or government programs that provide insurance, rural households receive some insurance through informal risk sharing networks and the use of assets. These are often effective strategies but are limited in the types of risk they allow households to insure against; they are also often very costly. Developments in index-based insurance and price insurance schemes, options to provide insurance to commodity farmers and facilitate producers’ access to future markets, and warehouse receipt schemes to provide insurance on staple crops are all ways to provide new tools by which this missing market can be addressed.
Finally, we would like to briefly comment on the authors' suggested statistical link between “political unrest” and price levels and volatility. We are very curious about how robust their regression results are to various changes in both the specification and estimation of their model. In particular, we would like to experiment with the following: a) the presence of a lagged dependent variable (article count) as one of the explanatory variables and the fact that the regression error is most likely correlated through time suggest an estimation procedure that is not OLS, as used by the authors; b) the fact that the dependent variable is a count (a variable that is always non-negative and non-continuous) suggests that standard statistical tests, as reported by the authors, may be misleading; c) as argued above, the fact that alternative measures of volatility may be more accurate than the rolling unconditional standard deviations used by the authors; and d) the fact that a longer time series of commodity prices may prove more adequate than FAO’s Food Price Index.
Carlos Martins-Filho is Professor of Economics at the University of Colorado at Boulder and a Senior Research Fellow at the International Food Policy Research Institute
Maximo Torero is Director of the Markets and Trade Division at the International Food Policy Research Institute
See more information on food price variability