WPS5460 Policy Research Working Paper 5460 Are Commodity Prices More Volatile Now? A Long-Run Perspective Oscar Calvo-Gonzalez Rashmi Shankar Riccardo Trezzi The World Bank Latin America and the Caribbean Region Poverty Reduction and Economic Management Department Economic Policy Unit October 2010 Policy Research Working Paper 5460 Abstract Soaring commodity prices in 2007 and 2008 raised breaks in volatility vary considerably across individual concerns that volatility was also rising, which would have commodities, cautioning against generalizations based implications for welfare and therefore for the design of on the use of commodity price indices; the three most public policy interventions. The literature focuses on significant breaks common to most commodities are the trends in commodity prices rather than their volatility two world wars and the collapse of the Bretton-Woods characteristics. This paper contributes by examining system; and structural breaks marking increased price commodity price volatility with a newly compiled volatility are followed by breaks marking declines in monthly panel dataset on 45 individual commodity volatility so that there is no upward or downward trend prices from the end of the 18th century until today. in volatility over time. The main conclusions are: the timing and number of This paper--a product of the Economic Policy Unit, Poverty Reduction and Economic Management Department, Latin America and the Caribbean Region--is part of a larger effort in the department to provide policy relevant research on topics of interest to the region. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at rshankar@worldbank.org and ocalvogonzalez@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Are Commodity Prices More Volatile Now? A Long-Run Perspective Oscar Calvo-Gonzalez, Rashmi Shankar, Riccardo Trezzi JEL No.: F14, C59, E32 Key words: International commodity prices, volatility, structural breaks The authors would like to thank David Jacks, Emily Sinnott, John Nash, Cesar Calderon, and participants at the Authors' Conference for the Regional Flagship Study on Commodities for comments provided at various stages during the preparation of this paper that have been invaluable to its development. All remaining errors are fully attributable to the authors. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. 1. Introduction and Motivation On July 11, 2008 the spot price of West Texas intermediate crude oil peaked at US$ 146 per barrel, exactly double the price a year earlier. Since then, oil prices bottomed at US$30 in December 2008, rebounding to around US$ 70 as of the time of writing. Such see-saw movements have not been confined to oil only. In 2008 large increases were also recorded for foodstuffs, metals, and commodities in general. The "commodity boom" was everywhere in the news and in policymakers' agenda. The concern was not only about the elevated price of commodities, but also that these prices had become more volatile.1 Some analysts, including the OECD, argued that this higher commodity price volatility seemed to be driven by structural determinants.2 The issue of rising commodity price volatility is policy relevant. Poor countries with production and trade structures concentrated on commodities are vulnerable to price swings. The perceived welfare effects of variable commodity prices have inspired public policy interventions in developed countries as well.3 Even the beneficiaries of higher commodity prices, such as farmers, have expressed concerns that higher volatility renders hedging mechanisms ineffective.4 Their complaint was that options had become so expensive due to elevated levels of volatility and hence risk, so that using options as a hedge was no longer financially viable. In general, a large number of policies ­ price supports, buffer stocking, and producer and consumer subsidies ­ have been rationalized on the basis that smoothing commodity price volatility away carries significant welfare gains. While there is a large literature focused on the trend in price level, stemming from the Prebisch-Singer hypothesis of a secular decline in the relative prices of primary products, 1 International Institute for Sustainable Development 2008 argued that "[...]in the past 30 years, there have been as many price shocks across the range of commodities as there were in the preceding 75 years" Boom or bust: how commodity price volatility impedes poverty reduction, and what to do about it", IISD, January 2008. 2 OECD-FAO (2008) Agricultural Outlook 2008-2017 discussed at length permanent factors that may increase volatility and underlined how Least Developed Countries could be negatively affected in an era of "high commodity prices and high price volatility," pg. 4. 3 For example, the EU's Common Agricultural Policy, the Canadian Wheat Board, and US implicit agricultural price support policies. 4 "Price Volatility Adds to Worry on US Farms," New York Times article on April 22, 2008. 2 volatility characteristics of commodity prices have attracted less attention. In this paper, we therefore address the following question: has commodity price volatility increased? We do this by exploiting a newly compiled unbalanced panel dataset on 45 individual commodity prices spanning the 1784-2009 period at a monthly frequency. 2. Literature Review Ever since Prebisch (1950) and Singer (1950), a vast literature has grown around the issue of identifying a secular trend in commodity prices. This research has been mainly concerned that commodity prices tend to decline relative to the price of manufactures with obvious implications for the primary producers.5 In contrast, the literature on the volatility of commodity prices and its public policy implications is relatively small. This research is mainly focused on understanding how commodity price volatility affects income volatility, especially in poor countries (Koren and Tenreyro (2007)), among others). Other authors have been concerned with how commodity price volatility generates instability in international markets (Blanford (1983), Heifner and Kinoshita (1994) among others). While mainly concerned with identifying price cycles ­ booms and slumps ­ Cashin and McDermott (2002) also test to see if variability in prices is higher or lower across cycles. Using The Economist's index of industrial commodity prices over the period 1862 ­ 1999, they find evidence of a ratcheting up in the variability of commodity price movements around 1899 and then again in the early 1970s.6 The finding of higher variability in commodity prices after the end of the Bretton Woods era is in line with hypotheses of a link between nominal (and real) exchange rates and the volatility of dollar denominated commodity prices Chu and Morrison (1984), Reinhart and Wickham (1994), Cuddington and Liang (1999). Comparing three different datasets7 5 Interest about a possible negative long-term trend in commodity prices has occupied development economic literature since the late 1940s. For instance in Prebisch (1950) and Singer (1950), Grilli and Yang (1988), Cuddington and Urzua (1989), Cuddington (1992), Powell (1991), Reinhart and Wickham (1994). 6 For annual data, Cashin and McDermott define large booms as a sequence of generally increasing prices that have had a price movement of at least 25 percent over the phase, and large slumps as a sequence of generally decreasing prices that have had a price movement of at least 25 percent over the phase. 7 The datasets are the following: (1) the annual data set of Grilli and Yang from 1900 to 1992, (2) Boughton's dataset with annual observations from 1854 to 1990, (3) the IMF's International Financial Statistics (IFS) which covers the post World War II period with monthly observations. 3 and using the methodology in Eichengreen (1994) to identify exchange rates regimes, Cuddington and Liang (2003) find that the relative price of primary commodities in terms of manufactured goods exhibits greater volatility since the early 1970s, a period characterized by an increasing number of flexible-exchange rate regimes. Mitchell (1987) presented the idea that increased trade, capital flows, policy shocks to macroeconomic variables and exchange rate uncertainty affect agricultural commodity prices. Moledina, Roe, and Shane (2004) find little evidence for higher volatility post-1971, once predictable components are removed from the "classic" measure of volatility (the unconditional standard deviation or the coefficient of variation8). Moreover, they show no statistical evidence for either a positive or negative trend in median volatility. Only three of the twelve commodities in their sample (bananas, coffee, and wheat) show some increase in volatility over the sample period. They argue that the welfare gains from eliminating commodity price volatility are therefore tiny, at less than one percent of total consumption. One lesson from Moledina et al (2004) is that the absence of a common trend across commodity prices calls for a need to study factors underlying commodity price volatility separately in each market. This was also highlighted by Leon and Soto (1995), who had analyzed the long-run dynamics of the price of the 24 most traded commodities over the 1990 ­ 1992 period. They tested for the presence of unit roots in the series, allowing for endogenously determined structural breaks. The results show that 15 of the 24 commodity prices in their sample exhibit a negative trend, six are trendless, and three show a positive trend9. Closest to this paper in spirit is Jacks, O'Rourke and Williamson (2009), who also examine commodity prices over a long time span. They define volatility as the standard deviation of price changes over a given period and use monthly observations on local market prices (to account for the impact of tariffs and embargoes) for four broadly defined commodity indices. They use the standard UNCTAD classification: all food 8 See Offutt and Blandford (1981) for a list of different single variable measures based on the standard deviation. Moreover, see Kroner, Kneafsey, and Claessens (1993) for a "classical" approach using the standard deviation of price changes. 9 The Cusum Break Test for TS models shows the following breaks: Coffee 1945, Maize 1920, Palm Oil 1985, Rice 1920, Sugar 1922, Timber 1985, and Tin 1985. 4 (AF), agricultural raw materials (ARM), minerals, ores and metals (MOM) plus a fourth group for manufactures or final good (FG). We provide below a review of the existing literature on commodity prices volatility. Author Research Question Data Methodology Results Moledina A., Roe The paper tries to determine Monthly prices (1957- GARCH (or ARIMA) The paper attributes most of L., Shane M. the most appropriate measure 2001) for selected model fitted to the volatility in commodity (2004) of commodity prices volatility agricultural compute conditional prices to macroeconomic and and investigates the presence commodities from volatility. political factors. Moreover, of a linear time trend in the IMF's IFS. CPI data the median volatility over conditional volatility series. from Bureau of Labor time for all commodities does Statistics are used to not show consistent increases deflate each commodity or decreases. series. Cuddington J.T., Do metal prices exhibit any Annual data series (in Band-pass filters Considerable evidence of Jerrett D. (2008) long-term upward or some cases from 1850) developed by Baxter three past super-cycles in real downward trends? Is there on the LME-traded and King (1999) and metal prices. Moreover it evidence of a strong super- nonferrous metals from Christiano and finds evidence for high cycle component for each Heap (2005). US CPI Fitzgerald (2003). correlation of the six LME series? Is there any strong co- (2006=100) used to metal super-cycles. movement of the super-cycle deflate the series. components? How pronounced are the shorter cycles? Jacks S., O'Rourke Has commodity price Monthly local city The paper presents No evidence of an increase of K.H., Williamson increased over time? Have prices (UNCTAD descriptive statistics commodity prices volatility J.G. (2009) commodities always shown classification) from of a selected measure though time. It also greater price volatility than different sources, of volatility. Prices concludes that higher manufactures? Does market including Bezanson volatility is defined as commodity price volatility integration breed more or less (2005), Posthumus the standard deviation compared to manufactures commodity price volatility? (1946), Friis and of the price ratio. has been a constant since 18th Glamann (1958), Gayer, century. It finds evidence Rostow, Schwartz that economic isolation has (1953), IMF and been associated with higher UNCTAD. commodity price volatility. 5 Cashin P., The paper addresses two Real annual data of the Following Watson No evidence of a break in the McDermott C.J. research questions: what is the nominal industrial (1994) the paper long-run trend decline in (2001) empirical behavior of commodities price index applies two commodity prices. Evidence commodity prices? Are there (dollar based), deflated econometric tests on a of a ratcheting up in the any changes in the variability by the GDP deflator of "peak-trough" variability of price of commodity prices and in the United States over analysis to determine movements. The amplitude the trend growth of prices over the period 1862-1999. the length of a cycle of price movements increased time? and its statistical in the early 1900s, while the significance. frequency of large price Descriptive statistics movements increased after complement the the collapse of the Bretton analysis. Woods regime of fixed exchange rates. Cuddington J.T., The study investigates The paper combines ANOVA tests are Strong evidence supporting Liang H. (2003) differences in real primary three datasets: Grilli and performed on long- the conjecture that the commodity price volatility Yang (1988), Boughton run data to test equal flexible exchange periods across fixed and flexible (1991) and IMF's IFS. variance across fixed have been associated with exchange rate regimes. The paper examines a and flexible exchange higher real commodity price real commodity price rate periods. Dummy volatility than the fixed index, defined as the variables in a exchange periods. ratio of the chosen (T)GARCH (1,1) nominal commodity framework are used to index deflated by a test the presence and manufacturing unit persistence of a value (MUV) index. volatility shock. Leon J., Soto R. The paper revises the long and Monthly prices of the The paper applies the 15 of the 24 commodity (1995) short run time series structure twenty-four most-traded non-parametric prices have negative trends, 6 of commodity prices in order commodities in 1900- estimator of are trendless and 3 exhibit to answer the questions of the 92. persistence proposed positive trends implying that secular decline and the long- by Cochrane (1988 Prebisch-Singer hypothesis run persistence of shocks. and 1991) and holds for most commodities. extended by Lo and Evidence suggests that there McKinley (1989) and may be substantial room for Chow and Denning stabilization and price (1993). support mechanisms for most commodities. 6 Here, the question of whether commodity price volatility has increased over time is addressed with analytical rigor and at a disaggregated level. We use a unique, newly compiled dataset of 45 individual commodity prices and five commodity price indices at a monthly frequency. Two econometric tests are applied for identifying structural breaks in GARCH-type processes in order to provide a robustness check and these are described below. 3. Data For this paper we use a newly compiled dataset which covers 45 individual commodities and 5 commodity price indexes. We use a monthly unbalanced panel of observations from Global Financial Data (GFD) covering the period 1784 ­ 2009. We take a pragmatic approach as to what constitutes a "commodity" and take all "commodities" included in the GFD dataset as our subject of study. We use data from GFD as it is the most comprehensive source, verifying that the price data is consistent with that from alternative sources (World Bank, IMF, and UNCTAD) for any overlapping commodity and time period. The details of the dataset can be found in Table 1. 7 Table 1. Availability of monthly price data for individual commodities and indexes Before proceeding to discuss potential tests for multiple structural breaks we need to further characterize the underlying time series data. The commodity price series show a high degree of persistence and a clear heteroskedastic component. This suggests the use 8 of a GARCH process to characterize our data. GARCH models, which have become extremely popular since their introduction by Engle (1982), allow for time varying volatility and permit the inclusion of additional structural determinants that can tell you how volatility is changing and what drives it. We formally tested the validity of this approach by fitting an AR(1) model and found the coefficient to be statistical significant at 1% level for all commodities (see results for selected commodities in Table 2). We then tested the presence of heteroskedaticity using a Breusch-Pagan ARCH Test on the residuals of the AR(1) regression and we found evidence of heteroskedasticity for all commodities as shown in the right hand side of Table 2. Visual inspection of the residuals from AR(1) regressions clearly exhibit volatility clusters as shown in Chart 1 for selected commodities. Overall, the choice of a (G)ARCH process to model the behavior of commodity prices appears robust. Table 2. AR(1) coefficients and Breusch-Pagan ARCH Test for selected commodities AR (1) Breusch - Pagan ARCH Test Coefficient t-Statistics F - Statistics Probability Coal 0.99 221.2 179.7 0.00 Coffee 0.99 283.1 32.4 0.00 Copper 0.99 229.4 176.8 0.00 Corn 0.99 336.2 59.0 0.00 Cotton 0.99 340.9 119.3 0.00 Crude Petroleum 0.93 156.5 87.5 0.00 Gold 1.00 743.7 92.4 0.00 Lead 0.99 384.6 250.9 0.00 Rice 0.99 245.7 232.0 0.00 Silver 0.98 264.8 145.3 0.00 Sugar 0.98 286.7 453.8 0.00 Tea 1.00 243.3 144.7 0.00 Wheat 0.99 326.7 149.1 0.00 Zinc 0.99 372.9 378.0 0.00 9 Chart 1. Residuals from AR(1) model on selected commodity prices Actual minus fitted values Copper Corn 100 1.5 80 60 1 40 0.5 20 0 0 20 40 0.5 60 80 1 100 120 1.5 Sugar Wheat 20 2.5 2 15 1.5 10 1 5 0.5 0 0 0.5 5 1 10 1.5 15 2 Source: Authors' calculations 4. Methodology Our search for an econometric methodology to detect breaks in commodity price volatility was based largely on three criteria. First, it ought to allow for the detection of multiple breaks. Second, the dates of the breaks should not be the result of an arbitrarily imposed choice but should instead be endogenously generated by the chosen methodology. Finally, given that commodity prices can be characterized as GARCH processes, methodologies specifically designed for ARCH-type of models would be preferred. On the basis of these three criteria we selected two tests: the Kokoszka and Leipus (KL 2000) test and the Inclan and Tiao (IT 1994) test. Both tests have been 10 applied in the literature to (G)ARCH-type models ­ see for example Hillebrand and Schnabl (2006) and Hillebrand (2005) for the use of the KL test and Granger and Hyung (1999) who applied the IT test to examine breaks in the absolute returns of the S&P 500. To explain the KL (2000) test it is useful to start from the basic GARCH definition. ARCH models are generally defined by two equations: rk k k , k2 a b( j )rk2 j (1) j 1 Where k are iid errors, and a and b( j ) are non-negative constants. These equations are suitable for series rk such that the observations are uncorrelated but exhibit clusters of volatility. Note that if k have zero mean, unit variance and the sequence rk is weakly stationary, then: Var rk a /(1 b( j )) (2) j 1 The KL test assumes that the parameters a and b( j ) change at an unknown point k * in such a way that the variance given by (2) changes. The test statistic is a CUSUM-type estimator k ^ of k * defined as follows: k ^ min k : Rk max R j 1 j n (3) Where k (n k ) 1 k 2 1 n Rk rj r 2 (4) 2 k nk j n j 1 j k 1 R ( k ) / ^ The normalized test sup k is asymptotically distributed as a Kolmogorov- ^ Smirnov process, where is an estimate of the long-run standard deviation estimated using a VAR HAC estimator.10 The general approach is to begin with the full set of observations to determine the first break. This break is then used to split the sample into two sub-series. The estimator then calculates breaks for the sub-series in order to 10 See den Haan and Levin (1997). For the choice of lags, we have followed the formula Lags = floor(4(T(/100)^(2/9))). 11 establish additional breaks. This iterative procedure is stopped once a break is found to be statistically insignificant. The IT test is based on a centered version of the cumulative sum of squares presented by Brown, Durbin, and Evans (1975). It is an algorithm that applies in principle to independent series and is designed to find a break in the (unconditional) variance with unknown location. Define Ck, the cumulative sum of squares of a series of uncorrelated random variables {a t } with mean 0 and variances t2 , t 1,2,..., T . as follows: C k t 1 at2 k (5) Also define Dk, k 1,..., T , with D0 DT 0 , as the centered cumulative sum of squares. Ck k Dk (6) CT T Time series with no breaks in variance would exhibit a plot of Dk oscillating around 0. A sudden change in variance would cause the plot of Dk to exhibit a pattern going out of the critical boundaries with high probability. Under variance homogeneity, the normalized statistic T / 2 Dk behaves asymptotically like a Brownian bridge. The advantage of the KL test is its validity under a wide class of processes, including long memory, GARCH-type and non-linear time series models. The relative advantage of the IT test is its simplicity and its independence from estimated long-run variance (which makes the test robust to time period selection). The Montecarlo simulations of Andreou and Ghysels (2002) suggest that the IT test has power and only minor size distortions when applied to strongly dependent data, though it is not as powerful as the KL (2000) test. It suffers from size distortions (above 10%) for all data generating processes (high and low volatility persistence) but appears to have good power in detecting even small changes in the GARCH coefficients or the error 12 process for large T. The test is not seriously affected by outliers for large samples (T > 3000). The KL test has good power only for large and non-monotonic changes in the GARCH parameters for any data generating processes for the absolute returns rather than the squared. The KL (2000) test shows good power for detecting changes in the variance of the error terms in the GARCH process and appears fully robust to outliers. Finally, as the sample size (T) increases the performance of the test improves even for small change points. We use both monthly and daily data on individual commodities in order to account fully for differences across commodity markets since both the mean and volatility in individual prices exhibits different break points. Additionally, the use of individual prices allows us to examine whether the relationship between prices of different commodities has changed over time. Further, a measure of how this relationship has changed permits an examination of the role of financial market integration in determining price dispersion or synchronization. Finally, we introduce robustness checks given the scale-dependence of measured volatility. 5. Results Both tests highlight the heterogeneity across individual commodities. Even in periods where volatility breaks are more common it is far from the case that all, or even most commodities, exhibit the same regime shifts. This can be illustrated by Chart 2 below, which shows the percentage of commodities in our sample for which we detect a break in any given decade. As Chart 2 highlights, in most decades the proportion of commodities that experience a break is relatively low (below 10-15 percent using the KL test and less than 50 percent using the IT test). Even in decades of very high volatility, like the 1910s, 1940s, and 1970s there is a number of commodity prices which did not exhibit breaks in volatility. 13 Chart 2. Prevalence of structural breaks in volatility identified by KL and IT tests Source: Authors' calculations Still, there are three periods where structural breaks in volatility are more common: the two world wars and the collapse of the Bretton-Woods system. This can be further illustrated by Chart 3 below, which also distinguishes between breaks marking increased price volatility and those that mark decreased volatility. Again, the results are shown for both the KL and IT test, with the expected differences in the number of breaks found (as the IT test is less stringent and thus more breaks are likely to be found). However, the overall pattern of breaks in volatility is broadly similar (with the exception of the 2000s, an issue to which we will return later). 14 Chart 3. Number of structural breaks in volatility IT test KL test Source: Authors' calculations The reason for why the two tests appear to give contradictory results in the 1940s has got to do with the deregulation of previously controlled prices in the aftermath of World War II. At that point there was a spike in volatility because prices had been kept constant during the war. This is picked up as a break in volatility by the IT test while it is not the case for the KL test, which is fully robust to outliers and is picking up the decrease in volatility that will characterize the Bretton Woods era. Focusing on the period since the 1970s, it is worth noting that the KL test still detects upward breaks in volatility in the 1980s, the 1990s, and the 2000s. This suggests an overall increase in volatility as these breaks add to those observed during the 1970s (generally speaking the breaks in the 1980s and 1990s detected by the KL test affect commodities which did not see upward breaks during the 1970s). So, even using the 15 stringent KL test, fifty percent of all commodities are now in a higher volatility regime than during the Bretton Woods period (using the IT test 37 out of 45 individual commodities are now in a more volatile regime than they were during the Bretton Woods period). It is important to stress that when upward breaks in volatility are not followed by downward breaks, volatility increases as a whole. This can help to explain why commodity price volatility is currently higher but the evidence on the number of structural breaks in the 2000s seems ambiguous depending on whether we use the KL or IT test. The difference in results for the 2000s between the two tests can also be related to the possibility that 2008 may still prove to be an outlier (and therefore the KL test is less likely to determine a break so close to the end-point of a time series). The very different sensitivity of the two tests can be illustrated by Chart 4, which shows the number of commodities that have exhibited 0, 1, 2, 3, or 4 breaks since the 1970s. Because the KL is particularly apt for a long-term analysis, focusing on its results may not be that informative when analyzing short-term trends. Chart 4. Number of breaks since the 1970s (by structural breaks in volatility) Source: Authors' calculations Drawing on the IT test we can conclude that there have been a number of upward breaks in price volatility in recent years. This is illustrated by Chart 5 below, which shows the average number of commodities that see a break in any given year (calculated using 5- year moving averages). In fact, the number of commodities that have seen such upward breaks in volatility in the 2000s is close to that observed during the 1910s, 1940s, and 1970s. This makes the recent period the decade with the fourth-highest increase in 16 volatility as measured by the number of commodities experiencing upward breaks (again using the IT test). However, the evidence suggests that structural breaks marking increased price volatility are subsequently followed by downward breaks in volatility so that there is no upward or downward trend in volatility over time. Chart 5. Breaks in volatility identified by KL and IT tests IT breaks ­ 5 years moving average KL breaks ­ 5 years moving average Source: Authors' calculations 17 6. Conclusions Our paper is motivated by three main questions. First, is there a pattern of volatility across commodities over time? Second, is there a long-run trend in commodity price volatility? Third, are there identifiable breaks in commodity price volatility? For example, do we see changes in volatility when the world went from relatively open (1820 ­ 1913) to closed (1914 ­ 1949) and then to open (1950-2009)? Drawing on a large unbalanced panel of monthly prices of 45 individual commodities and 5 commodity price indices we address these issues by implementing two tests for detecting structural change. Both tests have the advantage of not imposing a priori the dates of potential structural breaks. While both tests are based on CUSUM-type test statistics, they differ in how they establish the threshold above which a change in volatility is considered a structural break. The Kokoszka-Leipus test uses an estimate of the long-run variance of the entire time series to determine such a threshold level. As a result, it is a more stringent test resulting in fewer breaks and by and large only three periods in history are associated with breaks in volatility in most (but not all) commodities: the two world wars and the collapse of the Bretton-Woods system. The Inclan-Tiao (IT) test, in contrast, considers only the properties of the time series since the last identified break in volatility. As a result, threshold volatilities are re-defined every time a break is found. This results in less stringent thresholds and therefore more breaks in volatility ­ both upwards and downwards. Therefore, even though break-points may overlap, the IT test, constructed to be more sensitive, picks up more frequent shifts. However, the historical breaks coincide under both methodologies. The main conclusions are as follows. First, the timing and number of breaks in volatility vary considerably across individual commodities. This result cautions against broad generalizations and the use of commodity price indices to analyze changes in volatility. Second, the three most significant breaks common to most (but not all) commodities, are the two world wars and the collapse of the Bretton-Woods system. In recent years, however, there has been an uptick in price volatility in a number of commodities. During the last food crisis though, it is clear that volatility spiked, starting to rise before the actual increase in price levels especially for the most tradable commodities. However, the 18 evidence suggests that structural breaks marking increased price volatility are subsequently followed by downward breaks in volatility so that there is no upward or downward trend in volatility over time. 19 Annex ­ selected individual commodity price series Cocoa Monthly Price Monthly Returns (% change) and KL (2000) test detected breaks Detected Breaks by KL (2000) test: December 1861 (Increase in volatility), June 1880 (Decrease in volatility), December 1918 (Increase in volatility). 20 Coffee Monthly Price Monthly Returns (% change) and KL (2000) test detected breaks Detected Breaks by KL (2000) test: September 1985 (Increase in volatility). 21 Copper Monthly Price Monthly Returns (% change) and KL (2000) test detected breaks Detected Breaks by KL (2000) test: No breaks detected. 22 Crude Petroleum Monthly Price Monthly Returns (% change) and KL (2000) test detected breaks Detected Breaks by KL (2000) test: July 1884 (Decrease in volatility), May 1981 (Increase in volatility). 23 Platinum Monthly Price Monthly Returns (% change) and KL (2000) test detected breaks Detected Breaks by KL (2000) test: July 1948 (Decrease in volatility), December 1984 (Increase in volatility). 24 Rubber Monthly Price Monthly Returns (% change) and KL (2000) test detected breaks Detected Breaks by KL (2000) test: July 1909 (Increase in volatility), May 1934 (Decrease in volatility). 25 Silver Monthly Price Monthly Returns (% change) and KL (2000) test detected breaks Detected Breaks by KL (2000) test: October 1813 (Increase in volatility), July 1819 (Decrease in volatility), May 1862 (Increase in volatility), June 1866 (Decrease in volatility), October 1915 (Increase in volatility), November 1973 (Increase in volatility). 26 Soybeans Monthly Price Monthly Returns (% change) and KL (2000) test detected breaks Detected Breaks by KL (2000) test: May 1931 (Increase in volatility), September 1941 (Decrease in volatility). 27 Sugar Monthly Price Monthly Returns (% change) and KL (2000) test detected breaks Detected Breaks by KL (2000) test: June 1812 (Increase in volatility), May 1818 (Decrease in volatility), October 1855 (Increase in volatility), May 1865 (Decrease in volatility), June 1911 (Increase in volatility), January 1925 (Decrease in volatility), October 1962 (Increase in volatility), August 1988 (Decrease in volatility). 28 Wheat Monthly Price Monthly Returns (% change) and KL (2000) test detected breaks Detected Breaks by KL (2000) test: May 1870 (Decrease in volatility). 29 Zinc Monthly Price Monthly Returns (% change) and KL (2000) test detected breaks Detected Breaks by KL (2000) test: July 1914 (Increase in volatility), January 1917 (Decrease in volatility), August 1952 (Decrease in volatility), January 1973 (Increase in volatility), October 1974 (Decrease in volatility), January 2005 (Increase in volatility). 30 References Andreou, E. and Ghysels, E. (2002), "Detecting Multiple Breaks in Financial Market Volatility Dynamics", Journal of Applied Econometrics, 17:579-600. Blanford, D. (1983), "Instability in world grain markets", Journal of Agricultural Economics, 43: 379-395. Brown, R.L. and Durbin, J. and Evans, J.M. (1975), "Techniques for testing the constancy of regression relationships over time", Journal of the Royal Statistical Society, B 37, 149-163. Cashin, P. and McDermott, C.J. (2001), "The Long-Run Behavior of Commodity Prices: Small Trends and Big Variability", IMF Staff Papers 49(2): 175-99. Cuddington, J.T. (1992), "Long-run trends in 26 primary commodity prices: A disaggregated look at the Prebisch-Singer hypothesis", Journal of Development Economics, 39:207-227. Cuddington, J.T. and Liang, H. (2003), "Commodity Price Volatility across Exchange Rate Regimes", working paper. Cuddington J. T. and Jerrett D. (2008), "Super Cycles in Real Metals Prices?", IMF Staff Papers, 55(4): 541-565. Chu, K. and Morrison, T. K. (1984) "The 1981-82 Recession and Non-oil Primary Commodity Prices", IMF Staff Papers, 31: 93-140. Eichengreen, B. (1994), "History of the international monetary system: Implications for research in international macroeconomics and finance", in F. Ploeg, ed., Handbook of international macroeconomics (Blackwell, Oxford UK & Cambridge USA) 153- 197. Granger, C. and Hyung, N. (1999), "Occasional Structural Breaks and Long Memory," University of California at San Diego, Economics Working Paper Series 1999-14, Department of Economics, UC San Diego. Heifner, R. Kinoshita, R. (1994), "Differences among Commodities in Real Price Variability and Drift", The Journal of Agricultural Economics Research, 45(3): 10-20. 31 Hillebrand, E. and Schnabl, G. (2006), "A structural break in the effects of Japanese foreign exchange intervention on yen/dollar exchange rate volatility", Working Paper Series 650, European Central Bank. Inclan, C. and Tiao, G. C. (1994), "Use of Cumulative Sums of Squares for Retrospective Detection of Changes of Variance", Journal of the American Statistical Association, 89: 913-923. Jacks, D. S. and O'Rourke, K. H. and Williamson J. G. (2009), "Commodity Price Volatility and World Market Integration since 1700," NBER Working Papers 14748, National Bureau of Economic Research. Kokoszka, P. Leipus, R. (2000), "Change-point estimation in ARCH models", Bernoulli, 6: 513-539. Kokoszka, P. and Leipus, R. (1999), "Testing for parameter changes in ARCH models", Lietuvos Matematikos Rinkinys (Lithuanian Mathematical Journal), 39: 231--247. Kokoszka, P. Leipus, R. (1998), "Change-point in the mean of dependent observations", Statistics and Probability Letters, 40: 385-393. Leon, J. and Soto, R. (1995), "Structural breaks and long-run trends in commodity prices", World Bank Policy Research Working Paper 1406, January 1995. Miklos, K. and Tenreyro, S. (2007), "Volatility and Development" Quarterly Journal of Economics, 122(1): 243­287. Mitchell, D. O. (1987), "Factors affecting grain prices, 1970-84", Division Working Paper, World Bank, International Economics Department, International Commodities Markets Division. Moledina, A. and Roe, T. L. and Shane, M. (2004), "Measuring Commodity Price Volatility And The Welfare Consequences Of Eliminating Volatility", Annual meeting, August 1-4, Denver, CO 19963, American Agricultural Economics Association OECD-FAO (2008), Agricultural Outlook 2008-2017. Prebisch, R. (1950), "The Economic Development of Latin America and its Principal Problems", Economic Bulletin for Latin America, 7: l-22. PREM Notes (1999), "Using markets to deal with commodity price volatility", World Bank 13. 32 Reinhart, C. and Wickham, P. (1994), "Non-oil commodity prices: Cyclical weakness or secular decline?", MPRA Paper 13871, University Library of Munich, Germany. Singer, H. (1950), "The distribution of gains between investing and borrowing countries", American Economic Review, Papers and Proceedings, 40: 473-485. Singer, H. (1975), "The distribution of gains revisited", In The Strategy of International Development, ed. H. Singer, pp. 58-66. Macmillan Press, London. 33