Food Price Volatility

Transcription

1 Multi-year Expert Meeting on Commodities Palais des Nations, Geneva March 2010 Food Price Volatility by Christopher L. Gilbert University of Trento, Italy and C. Wyn Morgan University of Nottingham, UK "The views expressed are those of the author and do not necessarily reflect the views of UNCTAD"

2 Food Price Volatility Christopher L. Gilbert University of Trento, Italy, and C. Wyn Morgan University of Nottingham, UK, Presentation prepared for the Multi-year Expert Meeting on Commodities and Development, UNCTAD, Geneva, March 2010 based on work undertaken under the Foresight Project on Global Food and Farming Futures, Department for Business, Information and Skills (BIS), UK. The views expressed herein are those of the authors and not the BIS.

3 Introductory comments The food price spike has shocked governments and international agencies into a reassessment of likely developments in food availability and prices and of the adequacy of existing food security policies. These concerns have been underlined by the fact that food prices have not dropped back to pre-2006 levels and remain particularly high in many food-importing developing countries. There are two questions: 1. Will food prices be higher in the future than over past decades? 2. Will food prices be more volatile in the future than over past decades? Our concern is with the second of these questions.

4 1. Background

5 IMF commodity price indices Food prices remain 40% above 2005 values in nominal terms (18% above relative to advanced countries export unit values)

6 Grains prices Grains prices rose by more than other food prices with the (residual) free market rice price moving the most. Prices have fallen back through 2009 but the rice price is more than double its 2005 level (78% up in deflated terms) while maize and soybeans are 65% higher than in 2005 (40% in real terms).

7 Maize prices in Malawi Source: Authors calculations from data from GIEWSNET, FEWSNET, SAFEX and IMF. White maize is the staple food crop in most of East and Southern Africa. The Malawian maize price rose by much more than the inter-national price - here the South African futures exchange (SAFEX) price. It remains at a substantial premium to SAFEX. Similar situations exist in other East and Southern African countries.

8 2. Volatility

9 Volatility definition Volatility measures of the directionless extent of the variability of a price or quantity. It is measured by the second moment of an appropriate distribution, or transformations thereof. More frequently, economists measure price volatility as the standard deviation of logarithmic prices since this is a unit free measure. Many economic series exhibit trends. For such series, volatility is measured relative to the trend. Measurement of volatility therefore requires that the series be detrended. This requires a trend model and implies a judgemental trade-off between attribution of variability to the trend and to variation about the trend. Economists circumvent these issues by measuring volatility as the standard deviation of price returns, i.e. the standard deviation of changes in logarithmic prices.

10 Volatility measures 1. Historical volatility: typically the standard deviation of log price changes over a defined period. 2. Realized volatility: the standard deviation of log price changes within a defined unit, e.g. the volatility of daily log price changes over each month. 3. Latent measures: volatility is seen as having a slowly changing underlying level. 4. Implied volatilities: volatilities used by the financial markets in pricing derivative and other contracts over a defined future. In what follows we focus on historical volatility (1) taking the standard deviation of log monthly average prices over one year) but also use both a GARCH model and a structural components model to extract estimated latent volatilities (3).

11 Determinants of volatility (1) Price volatility is caused by shocks to the supply and demand curves S and D. The extent of price volatility depends upon p p The variability of supply and demand The slopes of the supply and demand curves. S S D A poor harvest (supply shock) shifts supply from S to S. Price rises from p to p.

12 Determinants of volatility (2) It is easy to show that the price variance depends on the sum of the variances of the supply and demand shocks divided by the sum of the demand and supply elasticities. It follows that price volatility will increase if either a) production or consumption become more variable; b) stock demand becomes more variable; or c) demand or supply become less price elastic. If we want to argue that there has been a sustained increase in food price volatility we need to point to (at least) one of these factors.

13 Possible drivers of recent volatility changes 1. Global warming may result in more variable weather conditions increasing the variance of supply shocks this is possible but currently available evidence is anecdotal rather than systematic. 2. Inventory levels have become lower, particularly in grains where governments have been encouraged to rely in trade rather than food reserves. Low stocks reduce supply elasticities in periods of shortage. 3. The use of grains, vegetable oils and sugar as biofuel feedstocks may increase the linkage between food prices and the oil price resulting in import of oil price volatility into food prices an effective increase in the variance of demand shocks. 4. Some governments, particularly in Asia, are putting greater effort into insulating their economies from external price shocks. This can impose a greater adjustment burden on the rest of the world (reduced supply and demand elasticities in the world market). 5. Futures market activity, which may have an effect similar to demand or supply shocks.

14 3. The long view

15 Volatilities from 1970 We look at the historical volatility of monthly average prices, as reported by the IMF, deflated by the US PPI, from 1970 to the present. We consider 19 food and beverage commodities: Softs : cocoa, coffee, tea and sugar. Fruit: oranges, bananas. Meats: beef, lamb. Grains: wheat, rice, maize (corn), sorghum, soybeans Vegetable Oils: coconut, soybean, groundnut, palm and sunflower oils. Fishmeal. We compare volatility in with

16 90s & 00s versus 70s & 80s 70% 60% 50% Only bananas and rice show a statistically significant volatility increase % 30% 20% 10% 0% Bananas Rice Sunflower Oil Coconut Oil Wheat Sorghum Coffee Maize Oranges Sugar Beef Cocoa Lamb Palm Oil Soybeans Soybean Oil Fishmeal Groundnut Oil Tea 8 out of 19 commodities show insignificant volatility change 9 out 19 food commodities show statistically significant (5%) volatility decreases

17 Summary of outcomes This set of tests indicate that food price volatility has generally been lower in the past two decades than previously. However 1. The results may be largely driven by the experience of the 1970s. 2. Grains are the most important food commodities. Their volatilities have not decreased. Rice volatility has increased. 3. There is evidence that volatility may have started to increase post-2005, particularly for grains.

18 4. The short view:

19 Was volatility higher in ? We look at this in two ways 1. GARCH model: We estimate a GARCH(1,1) model for each series of monthly volatilities and test whether the mean of the scedastic process was higher in Flexible Trend model: We estimate a smooth flexible trend for the annual volatilities analyzed previously up to 2006, and ask whether to volatilities are in line with those forecast by the model. The model incorporates a cycle allowing us to ask whether any recent rise in volatility is transient or permanent.

20 GARCH model ln pt =δ+ε t +θεt 1 ε N( 0, h ) h d h t t 2 t = µ +λ t +αε t 1+β t 1 This is a standard GARCH(1,1) model augmented by a dummy in the scedastic equation allowing a variance jump over The MA(1) specification takes into account serial correlation arising from use of average prices. d t is a dummy variable which is one from January The hypothesis test of interest is H 0 : λ = 0 against alternative H 1 : λ > 0, tested by the t statistic on λ (one-sided alternative so critical value is 1.65).

21 GARCH test results Maize 1.98 Coconut oil Rice 1.18 Groundnut oil 0.30 Sorghum 2.12 Palm oil 3.31 Soybeans 2.36 Soybean oil 3.88 Wheat 1.81 Sunflower oil 1.15 Cocoa 1.10 Sugar 1.10 Coffee Beef 1.07 Tea 0.82 Oranges Sugar: 1 st order process is ARMA(1,1). Significant positive coefficients for 4 out of 5 grain (rice is exception) and 2 out of 5 vegetable oils. No significance for remaining 6 food commodities. Implication: no general rise in volatility but evidence of some effect on grains and some vegetable oils. Bananas, lamb, fishmeal: model fails.

22 Flexible trend model ln vol = µ + c +ε µ t =µ t 1 +δt δ =δ +ν t t t t t t 1 t ct cosλ sin λ ct 1 κt * * * c =ρ t sin cos + c λ λ t 1 κt The model considers the logs of the annual volatilities lnvol where the volatilities vol are estimated as the standard deviation of monthly returns. µ is a smooth flexible trend with time-varying slope δ. c is a stochastic cycle which persists at rate ρ. The cycle frequency is defined by λ. We estimate over the sample and forecast We use a standard χ 2 (3) test to ask whether the forecast errors are compatible with the historical model.

23 Example: maize 40% 30% 20% 10% 0% Trend estimated to 2009 Trend estimated to volatility forecasts for Predictive failure test on one-step ahead forecasts χ 2 (3) = 2.32 (p-value 50.9%). Cusum t 3 = 1.37 (p-value 26.5%) forecast volatilities are substantially below realizations. However, because volatility is itself so variable the forecast errors are in line with the historically observed variability. Implications: maize price volatility was high but not exceptionally high. But the estimated volatility trend is now higher than in 2006.

24 Flexible trend results: grains and vegetable oils One step forecasts Estimated trend volatility χ 2 (3) Cusum t u/d 2009 Maize % 25.4% Rice % 17.9% Sorghum % 27.3% Soybeans % 25.9% Wheat % 21.1% Coconut oil % 22.4% Groundnut oil % 14.7% Palm oil % 33.5% Soybean oil % 19.5% Sunflower oil % 28.6%

25 Flexible trend results: other food commodities One step forecasts Estimated trend volatility χ 2 (3) Cusum t u/d 2009 Cocoa % 18.4% Coffee % 17.2% Tea % 25.5% Sugar % 21.6% Beef % 11.5% Lamb % 9.4% Fishmeal % 9.5% Bananas % 28.3% Oranges % 42.7%

26 Flexible trend results: summary 1. For all five grains and all five vegetable oils, trend volatilities are estimated as higher in 2009 than 2006, in come cases substantially so. (Note that the rise in estimated trend volatility could either result from an upward move in the trend or an upward revision of the earlier trend estimate). 2. For the other nine food commodities, there is no clear direction of change more estimated trend volatilities decline than rise. 3. Rice, sunflower oil and beef are signalled as having experienced exceptional volatility over

27 60% 50% 40% 30% 20% 10% 0% Estimated trend volatilities ranked by change Large volatility increases for major grains and some vegetable oils Large decreases for coffee, fishmeal and bananas Palm oil Rice Sorghum Maize Sunflower oil Wheat Beef Groundnut oil Soybeans Coconut oil Soybean oil Tea Cocoa Lamb Sugar Oranges Coffee Fishmeal Bananas

28 Conclusions 1. Food price volatility is itself very variable over time. Large price changes do not by themselves imply a change in volatility. 2. Food price volatility has generally been lower over the past two decades than in the 1970s and 1980s. 3. The volatility of grains prices does appear to have risen over the past three years perhaps because of lower stocks or through the biofuels link to oil prices. 4. The same appears to be true of some vegetable oil volatilities, in particular palm and sunflower oil. 5. Rice and beef both saw exceptional volatility in 2008, but in both cases this appears to have been transitory.

29 Thank you for your attention