Policy Research Working Paper 10531 Trading Places Fundamentals, Speculation, and Information in US Corn Markets Francisco Arroyo Marioli Development Economics Prospects Group July 2023 Policy Research Working Paper 10531 Abstract What explains the surge and plunge commodity markets typical seasonality seen in agricultural markets, incorporate have undergone in the past 20 years? Are speculators to supply and demand shocks as well as news shocks, and be blamed? Do prices reflect full information? These are allows for speculative storage decisions. The paper finds that the main questions addressed in this paper, in the context demand and supply fundamentals can account for around of the corn market. This paper formulates and calibrates 52 percent of past price changes from 1975 to 2016. The two quantitative models of corn prices formation. The first model also estimates the impact of information shocks to model is designed to explain prices in the long run (annual explain an additional 18 percent of quarterly deviations. frequency), while the second model applies to prices in the Finally, it finds that at least 30 percent of short-run price short run (quarterly frequency). For the long-run analysis, changes seem to have explanations other than supply or the paper finds that deviations of theoretical prices from demand fundamentals or information, demonstrating that observed ones are very small after 1996, and before 1996 when analyzing quarterly data, prices do not always closely they can be explained by government intervention. For track fundamentals. the short-run analysis, the model is designed to mimic the This paper is a product of the Prospects Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The author may be contacted at farroyomarioli@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 Trading Places: Fundamentals, Speculation, and Information in US Corn Markets Francisco Arroyo Marioli Key words: Information Acquisition, Market E¢ ciency, Commodities, Agricultural Demand, Agricultural Productivity JEL codes: G14, Q02, Q11 1 Introduction The objective of this paper is to study the fundamental and non fundamental determinants of com- modity prices over the past decades. The analysis speci…cally focuses on corn markets, for several reasons. Corn and soybeans are the main agricultural commodities in terms of production market value. They are the most traded contracts in futures market (next to cotton and wheat). Also, the US is the world leader in corn production, consumption and exports. It plays a signi…cant role in several states and sectors of the US economy, to the point where it has been considered a strategic sector for the US government for a long time. Prior to 1995, the US federal government would regularly intervene in these markets, usually seeking to guarantee minimum prices for producers by buying production. In more recent years, the government has shifted its intervention towards the demand side by incorporating biofuel mandates instead. Either policy signals this sector’s importance to policymakers even until today. This paper compares prices de…ned by fundamentals versus observed prices. It de…nes fun- damental determinants of price as demand and supply shocks. These may be either current or future. Future demand and supply shocks a¤ect current prices through information shocks (agents acknowledge that shocks will occur in the future) that change current inventory holding decisions. This paper is divided into two parts. The …rst part studies the current determinants of corn prices at an annual frequency, while the second part studies the short-run, cyclical determinants. The …rst section determines how annual prices would have changed solely based on current demand and supply shocks. The reason for this is simplicity: the paper shows that even when focusing only on current shocks, the model …ts the data very well. Any additional feature would only increase the performance of the model, thus making my results only stronger. The second part constructs theoretical prices at a quarterly frequency. Given that supply is zero in some quarters, the model introduces inventory purchases as part of the market. Given that inventory purchases are made within a pro…t maximizing scheme, expectations, and therefore future shocks through informa- tion, start playing a key role. The paper hence introduces in this section future shocks as part of fundamental-determined prices. More speci…cally, in the annual model, the paper analyzes the long-run price trend from 1975 to 2015. It then simulates a theoretical price time series, de…ned as the price such that each 3 s demand matches supply or, in other words, a price such that inventory variations are zero. year’ Moreover, it is even possible to assume that inventories could reach zero at the end of the year, given that, in practice, in the last quarter it usually reaches very low values with respect to harvest size. It proceeds to compare these theoretical prices with observed ones and estimate the di¤erence, and …nd that after 1996, di¤erences are small and do not last more than a year or two. Additionally, this document …nds that the increase in price levels in the 2000s seems to be explained mostly by the presence of an increasing demand for bioethanol. In the quarterly model, it follows a market structure that mimics the typical seasonality seen in agriculture markets ( with production occurring only during one quarter). Given that the time lapse between decisions is shorter, it also allows for information shocks to occur by introducing s expectations through private and public signals. Results show that around them to the agent’ 70% of price volatility can be explained through the shocks cited above, leaving a 30% measured space for other non fundamental sources of variation. Regarding model speci…cs, the long run case assumes isoelastic forms Dt = Zt pt for demand and St = At pt for supply, the …rst term represents (exogenous) levels and the second term the endogenous response to price with the respective elasticity. Therefore, shocks are captured as changes in Zt ; At . These will be de…ned as the current fundamental source of price changes. Time series for these variables can be obtained using USDA corn usage and production data. They are then integrated into a market clearing equation with no storage decisions. The short run case considers demand as a whole aggregate and assume identical isoelastic functions for both demand and supply, as before. The model is linearized, calibrated and simulated with real shocks as inputs. It then estimates the impact of di¤erent shocks (demand, supply, and information) on price changes and quanti…es them. One additional analysis that results from this paper is related to the issue of prices’being fully informative and appropriate for business-cycle measurement. Romer (2006) raises this question and shows evidence of …rms that do not always behave in pro…t-maximing ways by analyzing the case for professional football teams. This paper shows that although in the long run, prices seem to be very close to fundamentals, when going to a quarterly frequency, market prices can sometimes be very far away— as much as 50%— from them. These results are in line with Hussman (1992), who shows that once imprecise signals are introduced in a rational expectations framework, market prices become 4 ine¢ cient in transmitting information. Because corn markets are considered to be well-functioning markets, in the sense that prices and transactions are transparent, centralized, and very liquid, it is surprising and interesting to …nd that they might not always behave as one would expect in a classical supply-demand model with inventories and utility and pro…t-maximizing agents. This raises several relevant questions regarding macroeconomics, …nancial markets and industrial organization. First, if a very well developed market fails to deliver fully informative prices, what can then be said regarding other markets, where illiquidity, information asymmetry and search costs are more relevant? Moreover, since these results are found for quarterly data, the implications for business-cycle accounting could be important. In several developing countries commodity markets such as corn have a signi…cant impact on overall GDP and exports. If the prices that are being used for measurement are not market-clearing ones, important distortions could be taking place. Also, since annual prices do not di¤er from fundamentals as much as quarterly ones do, another potential question arises: Should we produce national accounts using prices measured at quarterly, annual, or some other frequency? How much time do markets need to become fully informative? Many business-cycle theorists believe that shocks on the real side of the economy, such as shocks to TFP and commodity prices, trigger and propagate economic ‡uctuations. Commodity price shocks clearly play important roles in developing nations today, particularly resource exporters and have also played a large role in economic ‡uctuations in the past. Economists and policy makers believed, for example, that the commodity-price declines in the late 1920s and early 1930s contributed to the length and severity of the Great Depression. In 1933, the Roosevelt administration’s e¤orts to raise commodity prices, particularly prices of farm products like corn, formed the centerpiece of its e¤orts to resurrect the economy. These results have implications for the sources of shocks and the accuracy of the interpretation of the shocks derived from RBC models. Given that productivity is a key driver of the cycle, a price system that does not track fundamentals closely could result in ine¢ cient resource allocations, or in other words, aggregate productivity losses. Therefore the implications of these …ndings are relevant regarding short-term business cycles. In the short run (at business-cycle frequencies), the majority of commodity price shocks do not re‡ect supply and demand, but could instead be a¤ected by speculative factors. These distortions could have a signi…cant impact on commodity producing economies. This paper is divided into four sections. The …rst section introduces and explains the main 5 structure of the paper and its relationship with the literature. It also applies a standard empirical approach used in the literature for oil markets to corn, to motivate and provide a benchmark for the models. Section 2 presents the low-frequency model and its results. Section 3 presents the high-frequency model, calibration, and estimation results. Section 4 summarizes and concludes. 1.1 Literature Review This paper, therefore, contributes to the literature in three respects: …rst, it tests the hypothesis that the spike in commodity prices was due to by non fundamental reasons, but focuses on a less explored market, since most of the literature has focused on oil markets. Second, it adds a new feature to high-frequency analysis: information shocks. This paper quanti…es the historical impact of information shocks by incorporating USDA reports as a source of information. Additionally, it estimates private signals and their impact on prices, showing that when it comes to quarterly analysis, they are a relevant source of volatility. Third, it decomposes price changes per source of change— that is, it estimates the impact of each variable change each year, from 1975 to 2015, and therefore o¤ers alternative explanations for the observed change in prices in past years. This also allows to estimate the fraction of prices that cannot be explained by fundamental or information factors, leaving a measured space for further research. Some papers have already attempted to use competitive models to study these markets. Roberts and Tran (2012) use a competitive model to estimate the role ethanol mandates in corn prices, …nding they can account for 11 to 0 percent of food price increase between 2005 and 2011. Zhou and Babcock (2017) use a model to estimate the role of hypotheically reducing biofuel mandates on prices. Mallory et al (2012) use this approach to link ethanol and corn markets, …nding that under certain conditions, future price expectations can in‡uence nearby futures across each other. Tegene et al (1988) make use of this method to estimate long-run price elasticity of corn acreage, …nding it to be 0.2. More generally, the literature has also examined whether commodity markets have been altered for non fundamental factors in the past decades, such as …nancial speculation. For instance, in energy markets results tend to indicate small or null e¤ects of speculation on oil prices. In their canonical paper, Kilian and Murphy (2014) develop a VAR model with speculative demand shocks and contrast it with recent oil inventory data and …nd no basis for speculation’s being blamed for 6 the 2003-2008 price period. They do, however, …nd it plausible that there was some in‡uence in previous years 1979, 1986, and 1990. Knitell and Pyndick (2016) analyze the oil market using a simple static partial equilibrium model with inventory markets and again …nd no relation between speculation and the oil price peak in 2008. Fattouh, Kilian, and Mahadeva (2013) summarize the literature that examines oil markets and conclude that there is no evidence that speculation is the main driver of price increases. Baumister and Hamilton (2019) also test a VAR approach for oil market but introducing priors for elasticities, …nding that supply disruptions tend to be a much bigger factor than in previous research. Another branch analyzes commodity …nancial markets from a portfolio point of view. Bohl and Stephan (2012) study the e¤ect of increased trading in future markets for six top traded agriculture and energy commodities using a GARCH model approach and …nd no evidence of a causal relation between future trading and price volatility. Chary, Lochstoer, and Ramadorai (2013) show that restrictions in …nancial markets can alter spot prices through hedging decisions, a¤ecting real outcomes. Sockin and Xiong (2015) study the possibility of information frictions in commodity markets, and demonstrate the importance of prices as signals for both demand and supply and the weakness of assuming that shocks are publicly known. In line with their results, this paper incorporates information shocks and allows for some shocks to be unknown. The work presented in this paper is mostly related to Knitell and Pyndick (2016), who estimate non fundamental shocks as deviations in inventory levels. The paper follows a similar methodology except that it consider corn markets (instead of oil) and a dynamic framework (rather than static). 1.2 A preliminary analysis s literature, though mostly for Empirical models have been used extensively in the commodity’ energy and metals. Given the lack of similar approaches used for corn, this section explores the use of these same techniques in corn markets The goal is to motivate the theoretical models used in this paper. The estimation proceeds to construct a VAR model with sign restrictions based on Kilian and Murphy’s (2014) model for oil. The endogenous variables are the same ones to be used later in the theoretical models: real price, consumption, production, and storage. Real prices, consumption, 7 and production are taken in log-di¤erences. Stocks are normalized as a fraction of same-year consumption. Sign restrictions follow Kilian and Murphy’s (2014) assumptions when applied to the same shocks: a positive supply shock increases consumption and production, while reducing prices; a positive demand shock increase prices as well as consumption and production; a positive speculative shock increases inventories, prices, and production, while decreasing consumption. The estimation uses annual data due to the extreme seasonality of corn production (three quarters of the year will always produce literally zero), and therefore add only one lag to model. Results can be seen in …gure I. The …gure shows the historical decomposition of price changes by shock. The …gure is quite indicative of the role of demand shocks: they clearly seem to play the biggest role, explaining 34 percent of price variation between 1981 and 2015. Speculative shocks explain 28 percent while supply explains 19 percent. This result holds some similarites to Kilian and Murphy’s (2014) model for oil, where they …nd that demand shocks are the biggest player in the long run, though in their case it is much more absolute almost 87 percent of the long-run real price ‡uctuations are due to demand. There are certain limitation though to this result. As the Kilian and Murphy mention in their 2014 paper: "Impact sign restrictions alone are typically too weak to be informative about the e¤ects of oil demand and oil supply shocks". Therefore, the next sections proceed to use theoretical approaches so to better identify the actual sources of corn price changes. 8 Figure I VAR Price Decomposition Figure I. Price changes decomposed by shock origin as estimated by the VAR model with sign restic- tions. Units are in percentage points. 2 Long Run Model The VAR approach is somewhat limited in its identi…cation capability. Both "speculation" and residuals are factors that are not so clear as to what they could mean economically.The goal in this section is to analyze and explain changes in price levels over a longer horizon by using a theoretical approach instead. In particular, one key question is whether observed prices have deviated from 9 fundamentals because of the presence of non fundamental factors, such as, due to the growing importance of …nancial speculation. This section proposes the following experiment: Calculate the theoretical price that would balance supply and demand every year and compare it with the observed price. That is, calculate the price such that inventory variation would have been zero. The intention behind this simple experiment is to see how prices would have changed solely based on current demand and supply shocks.Any additional features added here would only enhance the results, therefore the …t that will be shown can only be enhanced but not worsened by any additional feature of analysis we might want to add. This sections shows that even under this simple framework, results are very conclusive. It also implies an more long-run based approach. In the long-run, inventories cannot play a signi…cant role in price determination. Under the plausible assumption that speculators do not seek to buy-and-hold inventory, it is clear that in the long run prices cannot deviate far from fundamental ones. To understand this better, assume it is not the case. That is, assume that the fundamental price is systematically below or above equilibria with no inventory change. This would imply that stocks either decrease every year or accumulate in…nitely. Neither scenario is consistent with long-run equilibria, since it would either imply hitting the zero lower bound or an irrational accumulation of stocks. Since the analysis is long-run based this can be a useful approach. Formally, the theoretical price results from the following market-clearing equation: ! Zt pt = At p t : (1) f ood f eed ethanol exports Zt = Zt + Zt + Zt + Zt The left-hand side of the equation is demand for corn, and has four subcomponents: ethanol, food industry, feed (and residual) and exports. Supply has only one source: farming. Imports in the US are (practically) zero. For simplicity, demand and supply are assumed to have isoelastic functions of the price, as in Knitell and Pyndick (2016) and similar to Deaton and Laroque (1996). ! In this model, for demand functions, Zt ; captures exogenous demand shocks and pt the endoge- nous response to prices, with ! being the demand elasticity parameter. For supply, At represents 10 productivity and supply elasticity. h i 1 Zt !+ Equation (1) thus de…nes the theoretical market price as pt = At . Also, to quantify the impact of di¤erent shocks, one can calculate the …rst-order e¤ect for each of them. X @pt @pt i dpt = dZt + dAt ; i 2 ffood,feed,ethanol,exportsg (2) @Zt @At i where the right-hand side gives us the sum of the e¤ect of changes in all i 2 ffood,feed,ethanol,exportsg demand fundamentals and production fundamental A: USDA databases have time series for prices, all four demand uses (ethanol, feed, food, and exports), production, and yield per acre. The model uses these to estimate elasticities and identify exogenous shocks. For demand price elasticities, it uses instrumental variables, taking yield per acre shocks as a …rst-stage instrument for prices. Changes in yield per acre are explain mostly by weather factors. Farmers have almost no control over short run productivity levels. Therefore, it is reasonable to assume that they are exogenous to the production process and are appropriate instru- ments for prices. Supply elasticities are estimated following Roberts & Shlenker (2013). Estimation results can been seen in Table I.A. Results fall within other literature’s …ndings1 . Once elasticities are estimated, given demand, supply and price data, one can identify exogenous components by reversing the isoelastic equation and setting the exogenous component equal to the demand or supply level times prices elevated to inverse elasticity. i i ! Zt = Dt pt ; (3) At = St pt : i ; and p ; and estimated values for !; , one can decompose price Given time series for At ; Zt t changes per year by estimating the di¤erent terms of equation (2). As a result, changes in inventories explain the remaining residual. Results per variable can be seen in Figures II to V. 1 The Food and Agriculture Policy Institute shows elasticity estimations for corn in di¤erent countries, giving results that are always between 0.1 and 0.5. for both demand and supply. http://www.fapri.iastate.edu/tools/elasticity.aspx 11 The next step is to solve for the price equation (1) from 1981 to 2015. Figure II.A shows the theoretical price that results from the simulation and compares it against observed prices. The …gure suggests two key observations: First, di¤erences between the model-based and observed price are much larger before 1996 than after. That year, a more market-friendly US agriculture policy bill was passed, which reduced the budget for government purchases (hence the scope for intervention) and lowered price ‡oor targets below market equilibria. It is clear that after such an event, the distance between both time series reduces heavily. Regarding the 2000s spike in prices, the hypothesis of a non-fundamental driver seems pretty weak, since prices moved as one would expect them given demand and supply shocks. Second, theoretical volatility reduces signi…cantly after the year 1996. Table I.A Description Parameter Value Demand Elasticity ! 0:18 Supply Elasticity 0:15 Table I. *,**, and *** indicate p-values inferior to 0.1, 0.05, and 0.01, respectively: Table I.B Standard Deviation Variable All Prior to 1996 After 1996 Observed Price (in logs) 0:39 0:36 0:32 Fundamental Price (in logs) 0:59 0:74 0:36 Observed to Fundamental Ratio 0:41 0:59 0:12 Correlation Observed vs Fundamental Prices 0:82 0:72 0:96 Table I.B.The …rst two rows show standard deviations of both observed and fundamental prices for each time interval. The row "Observed to Fundamental Ratio" indicates the standard deviation of the ratio of observed prices divided by fundamental ones. The …nal row "Observed vs Fundamental Prices" shows the correlation between observed and fundamental prices for each time interval. A traditional statistical F test 12 was performed to check for a null hypothesis of equality in standard deviations of the ratio before and after 1996, results rejected the null hypothesis at a 1% level of signi…cancy. Figure II.A Theoretical Prices vs Observed Figure II.A. Theoretical prices are the values that solve the market clearing equation (1) given the identi…ed shocks. Observed ones are those obtained from the data. Units are in dollars per 100 metric tons of corn. Regarding annual variation, in summary, the model …nds that 60.4 percent of fundamental price changes were due to supply shocks and 39.6 to demand shocks. Non-fundamental shocks (inventory or speculative shocks) only play a major role prior to 1996, and almost none after. This is a sign…cant change relative to the VAR analysis, that assigns a smaller role to supply and much more to speculative shocks. The total role of supply and demand shocks is very robust to elasticities: 13 changing both supply and demand elasticities to values between 0.05 and 0.25 will change the total fraction of price changes induced by supply to be between 58 and 62 percent. In 1996, a Farm bill was passed. It signi…cantly shifted U.S. farming policy from a highly interventionist one toward a more "free market" approach. Prior to this bill, U.S. farming policy was heavily biased toward sustaining minimum price levels set at the discretion of the policymaker. The main tool through which this took place was by either government purchase of goods (The Commodity Credit Corporation - CCC- program) or subsidies to storage in the private sector (Farmer-Owned Grain Reserve program). Figure IV quanti…es the e¤ect these programs had on price levels. After 1996, the government budget allocated for purchasing goods was minimized, and minimum prices were set below market values; this resulting in a practically zero direct government intervention in corn markets. Table I.B shows statistical moments before and after the bill was passed in 1996. Table I.B shows standard deviations for observed prices and fundamental ones, standard deviation of the ratio and correlation. Two important facts emerge. First, the ratio between observed price and "fundamental" ones becomes signi…cantly more stable after 1996. A traditional statistical F test was performed to check for a null hypothesis of equality in standard deviations of the ratio before and after 1996, with results rejecting a null hypothesis at less than 1% signi…cance. This is also in line with an increase in correlation between both series. Second, the volatility of fundamental prices themselves also drop after 1996. This new stability could perhaps be one of the reasons for the lack of direct intervention during those years. Regarding the 2000s, it is clear that observed prices do not deviate too signi…cantly from fundamental ones. That is, the price that would have theoretically cleared the market with no storage decisions has not deviated signi…cantly from observed prices. This suggests that speculation (or other non fundamental shocks) has had little— if any— e¤ect on market prices. The immediate question is then: If not speculation, what has drives the price spikes? Appar- ently, as can be seen in …gure II.B, ethanol explains almost 56% of the price increase between 2005 and 2010. This can partially be explained by technological improvements (that allow ethanol to be used in energy markets) and government policy that has forced gasoline producers to incorporate mandatory fractions of ethanol in their …nal products. This decomposition has been calculated by inputting the estimated values of fundamentals and elasticities into equation (2). Additional evidence for this is the increased correlation between oil and corn prices, as in Figure VII (see 14 Appendix). Given that an increasing fraction of corn is used for ethanol, increasing correlation with energy markets is therefore to be expected, and this is what e¤ectively is seen in the data. This can be viewed as a new form of intervention, since instead of directly purchasing the product, they force private agents to use them. In future work, it would be interesting to use these results and calculate the impact of this intervention compared to previous ones, as well as other policy implications. In conclusion, the model used here suggests that there is very little ground for a nonfundamental explanation of price changes in levels during the past 20 years. More speci…cally, regarding the 2000s, the impact of biofuels on corn markets seems to be one of the main explanations for the observed behavior. Figure II.B Figures II.B. Fig. II.B shows the contribution of each demand factor to total price change (measured in percentage points change with respect to the previous year). They were calculated by replacing the 15 i ; A ,!; and estimated values of Zt ,i 2 ffood,feed,ethanol,exportsg ; into each term in equation (2). Each t term in equation (2) is represented by a shade of grey for a given year. Figure IV Figure IV. Fig. IV shows the contribution of total price change due to government CCC purchases. It was calculated by estimating the change in prices not explained by equation (2), i.e., the residual between explained price changes and observed price changes. That residual was then multiplied by the proportion of corn inventories held by the government under the CCC program. 3 High Frequency Case The previous model was designed to analyze long-run variation in prices. The main …nding is that there is little evidence of non-fundamental shocks. One may argue that this …nding is expected, since, as explained previously, a systematic deviation from our implied fundamental prices would mean either hitting the zero lower bound for inventories or increasing them to in…nity, as in Knittel and Pyndick (2016). However, at a higher frequency there could be space for short-term shocks 16 that are not necessarily related to supply and demand fundamentals. Therefore, though this section continues to assume isoelastic functional forms for supply and demand, it adds speculators as a new type of agent. The justi…cation for this is that given that production of corn occurs only during fall quarter , buying and selling storage inevitably becomes a relevant factor in determining market prices throughout the whole year. That is, give that production only occurs during on quarter, it is unrealistic to ignore inventory changes as a relevant factor. Since in a decentralized framework these inventory decisions are pro…t-seeking, the model rationalizes this by introducing a representative speculator: an agent who makes decisions based on expected pro…ts realized by buying low and selling high. The objective in the next section is to model short-run quarter equilibria and quantify the impact of both current and future shocks (the latter through information shocks) on total variation, similar to what was done in the previous section. 3.1 Model Time has two dimensions: year and quarter. Notation wise, t represents time change from quarter to quarter, q indicates a quarter-speci…c notation, and y represents marketing (not calendar) year. A marketing year starts with the harvest and ends exactly before the next one. For simplicity, demand is considered as a whole (that is, there is no di¤erentiation by use). Just as in the previous model, demand and supply are isoelastic. This seems reasonable, given that the demand elasticities found in the previous section were not so di¤erent across the various sources of demand. Dy;t (py;t ) = Zy;t py;t : Sy;t (py;t ) = Ay;t py;t : where Zy;t ; Ay;t are exogenous random variables representing demand and supply fundamentals and have the following dynamics: Ay;t = Ay = A Ay 1 + "A y if t = 1; Ay;t = 0, for t 2 f2; 3; 4g 17 Zy;t = Zy + "Z t : Zy = z Zy 1 + "Z y: Shocks are all IID and have the following distributions: Annual shocks "i y log N (0; i y) for i 2 fZ; Ag : Quarter-speci…c shocks "Z t log N (0; z q ); q stands for each quarter,1 to 4 . All shocks are summarized into vector 'y;t = "Z A Z y ; "y ; "t In other words, demand shocks have a yearly component that follows an AR(1) process, plus a quarter-speci…c noise. Production only comes during one quarter (harvest season); therefore, it is equivalent to de…ne supply shocks as annual or quarterly. In addition to farmers and con- sumers, speculators, participate in the market by buying and selling stored goods. They are pro…t- maximizing agents who face the following problem: Vt (xt ; xq ) = max pt ut f (xt ; xq ; ) + Et Vt+1 (xt+1 ) : ut ;xt+1 s.t. xt+1 = xt ut (xt xq )2 if t2f1;2;3g II f (xt ; xq ; ) = II ( xt xq ) 2 if t=4 > The function f (xt ; xq ; ) represents the cost of deviation from a certain quarterly optimum xq ; that has a speci…c value for each quarter and is known, i.e. it is neither random nor uncertain. II Deviating from this optimum implies some cost ; . The cost in the last quarter is di¤erent because usually inventories are managed at low levels during that time of the year. Therefore, given that agents are closer to hitting the zero lower bound, it is reasonable to assume that cost of deviating from optimum are higher. Intuitively, for some industries, having low inventories 18 could imply a huge risk premium: Since they require a minimum level of inventories to keep their machinery running, stopping them (by reaching zero inventories) would imply high costs. As an example, say that annual harvest usually has size 1, and quarter optimums are 0:8,0:6, 0:4; and 0:2; respectively. Having inventory levels below or above the …rst three does not imply the same costs as having them below 0:2, since in this last case companies would be close to hitting the zero lower bound. The speculator must decide one period in advance by how much stocks will di¤er from quarter- speci…c values. First-order conditions and envelope conditions are: h i pt + E Vt0+1 (xt+1 ; xq+1 ) = 0 h i @Vt (:) II 2 t+1 (xt xq ) + Et Vt0+1 (xt+1 ; xq+1 ) = @xt t+1 2 ; : Combing both, we get pt = E [pt+1 ] 2 t+1 (xt+1 xq+1 ) ; Policy function then results in: E [pt+1 ] pt Xt+1 = + xqt+1 : (4) 2 t+1 II with t+1 2 ; Consumers, farmers, and speculators must then meet at the market. Therefore, the market- clearing equation is given by: Xy;t+1 = Ay pt;y Zt;y pt;y + Xt;y if t = 1: (5) Xy;t+1 = Zt;y pt;y + Xt;y if t = 2; 3; 4: (6) 19 As a result, given that parameters and market-clearing conditions are quarter speci…c, equilib- rium prices will also be quarter speci…c. That is, their values will vary from quarter to quarter, even in steady state. 3.2 Information Structure Information is key to forming expectations for speculators, and therefore the information structure is a relevant issue in the model. This section constructs it to mimic reality as closely as possible. Production shocks and annual demand shocks are revealed the quarter after they take place, so agents do not know at present moments what supply and demand are. They use both public and private signals to form expectations regarding these. Private signals are formed each quarter. Public signals, on the other hand, are realized only in quarters 1 and 4. The model incorporates public signals this way in order to map USDA reports; hence, it must mimic the same timing schedule. Private signals are private in the sense that they are developed by the private sector. They are observable to speculators, but not to the econometrician. Information regarding demand and supply fundamentals is only revealed after each quarter is …nished. During each quarter, therefore the agent does not know with precision what is driving price changes. As an example, in quarter 1, yield, annual demand, and quarter-speci…c demand shocks are realized, but the agent only knows them after the quarter is over. Therefore, in quarter 2 the annual component of demand is known, but not the quarter-speci…c one for quarter 2, and so on. Uncertainty about current shocks takes place only in quarter 1, since in quarters 2 ,3 and 4 they only speculate with respect to next years fundamentals (which by de…nition have not taken place yet). So, in quarter 1, speculators can observe the current price, but since the current price will depend on the realization of three shocks (annual supply shock, annual demand shock and quarter 1 speci…c demand shock), they cannot identify or map back to fundamentals (there is no 1 to 1 mapping possible). The assumption used here is that if there is any information that can be captured from this it is negligible and for simplicity will be disregarded. Agents have access to public reports that are issued in the …rst and last quarters of the marketing year. The …rst-quarter report predicts current annual demand and yield and the latter predicts the upcoming year’ s yield and demand, on the s amounts. Private forecasts regarding next year’ 20 other hand, are made available every quarter. To summarize, public reports predict current yield (if quarter 1), upcoming yield (if quarter 4), current annual component of demand (if quarter 1), and upcoming demand annual shock (if quarter 4). To illustrate this, consider the case of a commodities trading company. During the …rst quarter, it receives the USDA report and the one it privately issues regarding current demand and supply conditions. Once the …rst quarter has passed, the company observes what happened in quarter 1, and its research department continues to develop forecasts for the upcoming year. In contrast the USDA will only make its own forecast once quarter 4 is reached, and so on. Table II Public Signaling Timing Formally, the speculator receives public and private signals at each period t : In the …rst quarter (or "harvest" quarter) t = 1, signals contain the following structure: shock being predicted noise z}|{ z }| { Apu pu;A y;t = "A y + t public Yield signal Zpu pu;Z y;t = "Z y + t public Demand signal Apr pr;A y;t = "A y + t private Yield signal Zpr pr;Z y;t = "Z y + t private Demand signal In non-harvest quarters t = 2; 3; 4 signals are: shock being predicted noise z}|{ z }| { Apu pu;A y;t = "A y +1 + t public Yield signal Zpu pu;Z y;t = "Z y +1 + t public Demand signal Apr pr;A y;t = "A y +1 + t private Yield signal Zpr pr;Z y;t = "Z y +1 + t private Demand signal 21 pu;A pu;Z pr;A pr;Z where t ; t ; t ; t are IID noise components and have a normal distribution with pu pu pr pr standard deviations (or inverse precision) A ; Z ; A ; Z . n o Apr Zpr Apu Zpu All signals are summarized into the vector y;t = y;t ; y;t ; y;t ; y;t The predicted variable will vary depending on the quarter. In harvest quarters, signals predict the annual demand shock and annual supply shock, which are taking place currently. In non- s annual demand shock and annual supply shock. This harvest quarters, signal predict next year’ can be seen more clearly in Table III. Therefore, each quarter there will be several sources of shocks: annual demand shocks, annual productivity shocks, quarter-speci…c demand shocks, and public and private information shocks regarding both demand and supply. Table III Predicted variable Harvest Quarter Non-Harvest Quarters Signal Quarter 1 Quarter 2 Quarter 3 Quarter 4 Apu y;t "A y - - "A y +1 Zpu y;t "Z y - - "Z y +1 Apr y;t "A y "A y +1 "A y +1 "A y +1 Zpr y;t "Z y "Z y +1 "Z y +1 "Z y +1 Table III.This table shows what variable is being predicted by each signal in each quarter. In the harvest quarter, signals intend to predict current shocks, since these not inmediately observable. In the rest of the year signals intend to predict next year’s shocks. Once a signal is received, agents use it to update their previous beliefs. That is, they use past private (public) signals and combine them with their latest private (public) signal, constructing a posterior private (public) signal. For example, in quarter 2, the private demand signal the agent receives is the …rst one that forecasts next year’s fundamentals. In this case she has a ‡at prior, and therefore her posterior will be identical to the signal received. In the third quarter, however, Zpr Zpr the agent has a previous private demand signal 2 and receives 3 : Assuming both signals have equal precision, she weights them optimally by assigning equal weight to both (recall that the noise 22 Zpr process is IID). Therefore, in the third quarter, posterior signal 3 will be obtained by following a Bayesian updating process with normally distributed noise: Zpr Zpr Zpr pr 1 pr 3;y = 0:5 2;y + 0:5 3;y with standard deviation 3;y;Z =p Z : 2 In the fourth quarter, she receives another private demand signal, which she again uses to update: Zpr 1 Zpr 1 Zpr 1 Zpr pr 1 pr 4;y = 2;y + 3;y + 4;y with standard deviation 4;y;Z =p Z : 3 3 3 3 In the …rst quarter the agent receives the last private demand signal Zpr 1 Zpr 1 Zpr 1 Zpr 1 Zpr pr 1 pr 1;y +1 = 2;y + 3;y + 4;y + 1;y +1 with standard deviation 1;y +1;Z =p Z : 4 4 4 4 4 For private signals regarding supply, constructions are identical. In the case of public signals, calculations are also identical, but since they only take place for two quarters (4 and 1), they have pu p1 pu pu p1 pu …nal precision A ; 2 A for supply and Z ; 2 Z for demand, respectively, for each quarter. Agents therefore have, each quarter, …nal public demand and supply signals, and …nal private demand and supply signals. To rationally summarize this information, these …nal signals are weighted as a function of their posterior precision. Formally, in quarters 1 and 4, they have Apu Zpu posterior public signals y;t ; y;t with a noise process that also follows a normal distribution, with pu 1 pu 1 precision A ; Z calculated above: They then update their beliefs according to Bayes’ rule, that is, they construct a single terminal signal for demand and a single terminal signal for supply, based on a precision-weighted average of both posterior private and public signals: A Apu Apr Apu Apr pu 1 pr 1 y;t = t;A y;t + 1 t;A y;t y;t ; y;t with precisions t;y;A ; t;y;A : Z Zpu Zpr Zpu Zpr pu 1 pr 1 y;t = t;Z y;t ; + 1 t;Z y;t y;t ; y;t with precisions t;y;Z ; t;y;Z : 23 pu 1 pu 1 4;y;A 4;y;Z 4;A = pu 1 pr 1 4;Z = pu 1 pr 1 for quarter 4, 4;y;A + 4;y;A 4;y;Z + 4;y;Z pu 1 pu 1 1;yA 1;y;Z 1;A = pu 1 pr 1 1;Z = pu 1 pr 1 for quarter 1, 1;y;A + 1;y;A 1;y;Z + 1;y;Z Parameters t;A ; t;Z t 2 f4; 1g indicate the weight agents put in each variable on quarters 4 and 1. Such weights depend on the relative precision of each …nal signal. De…nition 1 Speculators do not observe current A and Z ; instead they form expectations based on private and public signals regarding demand and supply. Hence, their information set y;t can be formally de…ned as: 8 9 > < > = For quarter t = 1: y;t = Ay 1 ; Zy 1 ; "zy 1 ;4 ; Xy;1 ; A ; Z 1; y 1 ;4 : > : | {z } | {z } | {z } |{z} | y;1{z y;} > ; annual productivity demand level previous quarter demand shock inventory 8 9 signals > > Ay ; Zy ; "z ; > > > > y;t 1 > > > < |{z} |{z} | {z } > = annual productivity demand level previous quarter demand shock For quarters t = 2; 3; 4: y;t = : > > A Z > > > > X y;t ; ; ; y;t 1 > > > : |{z} | y;t{z y;t } > ; inventory signals Therefore, expected prices can be de…ned as: E (py;t ) = E (py;t j y;t ): y;t 2 ; where is the set that contains all information and demand and supply shocks, and therefore contains y;t 8 fy; t:g Note that by construction, y;t 2 y;t ; 'y;t 1 2 y;t 8 fy; t:g and information sets are a combination of current signals and past shocks, that come from their own distributions. Therefore is a measurable space and y;t ; 'y;t 1 2 f ; F; P g The solution to the equilibrium system of equations will have the following functional— nonlinear— form,given that although speculators have a linear-quadratic objective, the rest of the supply and demand functions are isoelastic: py;t = py;t ( y;t ; 'y;t ) = p( Ay ; Zy ; "z y;t ; Xy;t ; A ; Z ; Ay 1 ; Zy 1 ): (7) |{z} |{z} |{z} |{z} | y;t{z y;t } productivity demand level quarter shock Inventory signals 24 De…nition 2 An equilibrium consists of prices py;t ( y;t ; 'y;t ) such that, given shocks 'y;t ; y;t , information set y;t , and initial conditions X0;1 ; A 1; Z 1, markets clear for all quarters and the speculator’s FOC is satis…ed for all fy; tg Prices will depend not only on the current level of productivity and demand, but also on the level of initial inventories and available signals. Past productivity and demand can in‡uence this by providing information about future realizations of demand and supply, since they follow an AR(1) process. Given this solution, two main issues remain to be addressed: First and most important, the solution is nonlinear, which makes it impossible to …nd an explicit solution. Second, private signals are non-observable in the data. Therefore the model will be solved by proceeding through the following steps: Step 1 Linearize the system around steady-state values. Normalize these to 1, such that index points can be interpreted as being close to percentage changes. Step 2 Calibrate parameters using USDA corn market data for 1975-2016. Step 3 Estimate seasonality and simulate the model by feeding estimated shocks as inputs, the compare model prices versus data. Step 4 Estimate private signals using a Kalman Filter. Step 5 Estimate the contribution of each factor to …nal prices and measure the residual (unexplained fraction of price changes). 25 3.2.1 Step 1 The …rst step is to linearize the model around steady state. Variables are rede…ned in the following way: For any variable y; ^=y y y where y is the steady-state value of the respective variable ^ is the linear deviation from it. and y Therefore, one must solve for steady-state values. Steady-state estimates are obtained in the following way: First, de…ne a harvest size of one: A = 1: Then normalize demand for each quarter, such that they all add up to the harvest 4 X Zi = 1: i Next, estimate quarter-speci…c seasonality Zi for demand in the data and normalize it such that they all add up to the harvest size: 0 Zi Zi = : X 4 Zi i 4 X 0 Therefore, one will have Zi = A = 1: i For inventories, in steady state they should not deviate from xq since the values for these are picked such that sticking to the target is always optimum. If speculator’s match their target xq , the marginal inventory cost is zero These are calibrated by taking average inventory levels (relative to harvest size) for each quarter throughout the whole sample. Therefore, the policy function equation s problem will give us the evolution of prices throughout each quarter in steady from the speculator’ state: Xt+1 xq+1 = 0 ) p 4 = 1 p3 = 1 2 p2 = 1 3 p1 : 26 Therefore, one only needs the value of prices in steady state for one quarter to determine the 0 rest. Given values A; Zi ; Xi ; i = 1; 2; 3; 4 and the price evolution path, one can introduce these values into steady-state market clearing equations to obtain p1 numerically: 0 X2 = Ap1 Z1 p1 + X1 : 0 X3 = Z2 p2 + X2 : 0 X4 = Z3 p3 + X3 : 0 X1 = Z4 p4 + X4 : Once steady-state values are calcuzalated, one can proceed to linearize previous equations (4) and (5). Linearizing the market-clearing equation results in: ^ t+1;y = Ay p 1 p X ^ ^t;y + Z p 1 ^ t;y : t;y ^t;y + pt Ay pt;y Z t;y ^t;y + X p (8) s policy function will then become: Speculator’ E [pt+1 ] pt t+1 Xt+1 = 2 t+1 : ) 2 Xt+1 = E [pt+1 ] pt 2 t+1 ^ t+1 = X ^0 E p t+1 ^t p (9) t+1 II with 2 ; : Therefore, variables will re‡ect deviation from steady-state values in levels. It is important to point out that steady-state values are quarter speci…c. Equations (8) and (9) are linear in prices and inventories. Therefore, the price solution equation will have also a linear solution. Since both 27 equations depend on parameters that vary in each quarter, the price solution equation will also vary depending on the quarter. Conclusion 3 As a result, the linearized new state space will have four price solution equations, one for each quarter: p ^1 = ^ IA + I "Z + ^ IX + I A + I Z + ^ IZ + ^ IA + ^ IZ 1 1 2 1 3 1 4 1 5 1 6 y 1 7 0 8 y: p ^2 = ^1 II A + II "Z + ^1 II X + II A + II Z + ^y 1 II Z + ^0 II A + ^y : II Z 1 2 1 3 4 1 5 1 6 7 8 p ^3 = ^1 III A + III "Z + ^1 III X + III A + III Z + ^y 1 III Z + ^0 III A + ^y : III Z 1 2 1 3 4 1 5 1 6 7 8 p ^4 = ^1 IV A + IV "Z + ^1 IV X + IV A 1 + IV Z 1 + ^y 1 IV Z + ^0 IV A + ^y : IV Z 1 2 1 3 4 5 6 7 8 A Apr Apu Z Zpr Zpu i =[ i i ] i =[ i i ]: where I; II ; III ; IV / for each state are price policy function parameters with i = 1; ::; 8 i i i i variable and q = I; ::; IV for each quarter. Recall that in steady state i = 0 for all signals , both private and public. It is important to remember that the sole role of "last year’s" production and demand is to inform about future production and demand. Therefore, if persistence parameters A and Z are equal to zero, J and J should also be zero for all J: 7 8 3.2.2 Step 2 The model uses USDA data series for yield per acre, production, demand (local and exports), prices, and stocks. The section proceeds …rst by detrending and deseasonalizing them, then normalize their mean so that index points can be read as percentage points. To obtain the exogenous demand component, the model instruments prices with yield per acre to obtain demand elasticity through a typical IV analysis. The next step is to use residuals as estimates for Zy;t ; just as in the long- run-trend model. Then, the section estimates the annual component of the residuals to obtain Zy s (2013) results. and Zt series. For supply elasticity , again the model uses Roberts & Shlenker’ As a robustness check, other values for were used with very similar results. 28 When it comes to reports, one can make use of digitalized USDA forecasts for each marketing Zpu year, from 1975 to 2016. These forecasts are for demand and yield per acre. To obtain t forecasts from demand forecasts, we regress these last ones against forecasted yields, then use the residual as a demand shock estimate. The reason for this is because given that a simple demand forecast contains also an endogenous forecast, since a higher yield would endogenously result in lower prices and higher consumption, one can eliminate this endogenous component of the demand forecast by regressing it against the yield forecast. This section calibrates the parameters to simulate and show quantitative results. Given the data available (described above), the section proceeds as the following. First, calibrate the parameters q I II necessary to calculate i which are ; ; ; A; Zy ; Zq ; !; and : Demand and supply elasticities are estimated as indicated in the previous step. The persistence of supply and demand fundamentals are calibrated by estimating AR(1) processes, and volatility is obtained by taking the residual’s standard deviations. I II Storage cost parameters ; are estimated by choosing values that minimize residuals. That is, the previous set of equations will give theoretical prices as the result for each quarter. The I II di¤erence between these and observed prices are the residual. Parameters ; were calibrated such that the total sum— in absolute values— of the residuals is minimized. The next step is to calibrate public and private signaling parameters. Regarding public signals, it is important to say that these appear in speci…c days, which are publicly known in advance. That is, at a certain time, information becomes publicly known in a report and markets react almost instantaneously Price changes for those "announcement" days can be seen, and even though private information is always present in markets, it is reasonable that by constraining price changes to the speci…c announcement day isolates public signal’s e¤ect on prices. The precision of public forecasts is estimated by comparing USDA demand and supply forecasts versus ex post realizations. The most non-straightforward task, however, is to obtain a value for the precision of private signals. The section proceeds to identify them as follows. First, calculate I and I by using the method of undetermined coe¢ cients (also use this to obtain all parameters 4 5 q i ). Second, obtain the release dates for public signals (USDA forecast reports) and the same-day price reactions. Then, regress same-day price changes against the changes in forecasted values in the released reports— with respect to the previous report— for that day. This allows to estimate the 29 e¤ect of public signals on price changes. The parameter that relates price changes to public supply signaling in the model, 4 1;A ;should match this previous price reaction estimation. Once known the value for 4 1;A , and given that also 4 ;is known one can obtain the value of the weight of pu 1 1;yA public signals relative to private ones, 1;A Since 1;A = pu 1 pr 1 and public signal precision, 1;y;A + 1;y;A pu pr pr A ; is known, one can identify private signal precision, A . The procedure to identify Z is identical. Calibration results for US corn markets can be seen in Table IV.A: Table IV.A Description Variable Value Source Discount factor 0:995 Literature Demand elasticity ! 0:18 IV Supply elasticity 0:15 Literature Demand persistence z 0:64 AR(1) Zt Supply persistence A 0 AR(1) At Steady State productivity A 1 Normalization Steady State demand level Zy 0:25 Normalization I Inventory adjustment cost quarters I,II,III 0:0204 Model calibration II Inventory adjustment cost quarter IV 0:046 Model calibration Yield volatility A 0:098 Stand Dev At Demand volatility Z 0:018 Stand Dev Zt pu Public signal A st dev A 0:06 USDA Yield forecast pu Public signal Z st dev Z 0:048 USDA Demand forecast 3.2.3 Step 3 30 Once determined the values for the parameters, the model is simulated by feeding in shocks ob- tained from data. That is, the data is normalized, detrended, and deseasonalized, while and use the residuals as supply, demand, information, and inventory state variables. Since one cannot observe private signals, for now they will assumed to be zero (they will later be estimated). Simulation results can bee seen in Figure VIII. Figure VIII Figure VIII. Lines show the evolution of observed corn prices versus model simulated ones. Units are deviations from steady state in percentage points. Prices Figure VIII shows the results given by the simulation versus those in the data. At …rst glance, one can say the model generates a price series that is in line with the one observed in the data. However, it is clear that for some years there are discrepancies between model-generated and observed prices. This can be better seen in Figure IX, which shows residuals measured as di¤erential between observed and e¤ective prices. On average, around 36% of price changes cannot be explained within the model. However, this is not uniform throughout time. Indeed, there are quarters, such as 1985-1987, in which non-modeled factors explain almost 60% of price changes, and 31 others in which that drops to less than 10%. There are several reasons for this. From a theoretical point of view, the model is a linear approximation to equilibrium values. Hence, second-order e¤ects might be bigger than expected. From a practical perspective, there could be non-pro…t-maximizing institutions in place— for example, government policy. Agriculture is an industry subject of several policies, with tax breaks, subsidies, inventory, and price policies in all world markets, with the US being no exception. It is important to say, however, that unexplained factors do not seem to have increased in the past 15 years. Figure IX Figure IX. Bars show the percentage points of price changes that cannot be explained by factors in the model. For some periods of time, prices can di¤er almost 60% from these based on their fundamentals. 32 3.2.4 Step 4 A key innovation of the model is the incorporation of both private and public signals into agents’ decisions. Agents use these signals to predict the annual component of demand and yields. They have both private and public sources, and weight them proportional to their precision. The model makes use of USDA reports released monthly that predict future demand and yield per acre; they are considered be public signals since anyone can access them at no cost. The section then estimates private signals through a Kalman Filter, in which the observables are prices and ex post realized annual demand and yield values. Formally: xt = Ft xt 1 + "t : state space model yt = Ht xt + Bt ut + t: observables where yt = [pt "A Z t+1 "y +1 ]: Apu Zpu ut = [At "Z t Xt t t Zy 1 Zy ]: Apr Zpr xt = [ t t ]: and " are covariance matrices for t ; "t : is calibrated by measuring the precision of public forecasts and the standard deviation of changes in price that remain unexplained by the model. Values for " ;are the precision in private forecasts estimated previously. Matrices Ft ; Ht ; and Bt represent the laws of motion for private signals, the e¤ect of private signals on prices, and the e¤ect of the rest of state variables on prices, respectively. Also, the second and third rows of matrix Ht represent the relation between signals and ex post realizations. That is, they capture the fact that signals forecast ex post realized demand and supply shocks with some error. Estimations can be seen in Figures X and XI, which show the estimated private signal. Although private and public signals tend to move together, as one would expect, there are several moments in which these di¤er, which may explain public information is not the only driver of expectations. 33 Figure X 34 Figure XI Figures X - XI. Bars show the forecasted demand and supply shock for each quarter by both private and public signals. Fig. X shows supply shock forecasts and Fig. XI shows demand shock forecasts. Darker bars show forecasts from USDA, and light bars show private forecasts estimated using a Kalman …lter. An important step is not only estimating signals, but also to checking if they are relevant. Figures XII and XIII (see Appendix) show the contribution to price change in index points for 1975-1979. In certain quarters, their impact on prices can reach almost 10%. That is, these information shocks are economically signi…cant the moment they occur. When considered over the whole sample time span, their relevance di¤ers based on whether they are public or private. In the …rst case, overall impact drops to 2.6% (combining demand and supply forecasts), while in the latter it reaches 15%. This is not surprising, since in the model public signals only occur in two quarters each year, whereas private signals take place every quarter. Also, private signals are more volatile, and therefore have a greater impact in total volatility. 35 3.2.5 Step 5 Table IV.B describes the contribution of each factor to price changes, measured by the average ab- solute value of factor e¤ect over price change (according to model policy parameters). As expected, the main driving factors are demand components (both annual and quarterly) and inventories, since when they are combined we reach almost 50% of price variation. The number associated with yield per acre is particularly small, but this should not surprise since the model divides time into quarters, with production (yield) only in one of four quarters. The supply in‡uence on prices, hence, tends to be absorbed by stocks in the following quarters, since when there is a good or bad yield this later changes the level of stocks during the rest of the marketing year. Figure XIV Figure XIV. Decomposition of price changes per factor. Di¤erent shades indicate the individual e¤ect of each variable over total price change for that quarter. Units are in percentage points. 36 Table IV.B Fraction of Price Change Due to Each Variable* Variable Description Fraction Apr t Supply Private Signal 0:056 Zpr t Demand Private Signal 0:099 At Productivity 0:018 "Z t Quarter Demand Shock 0:089 Xt Inventories 0:088 Apu t Supply Public Signal 0:009 Zpu t Demand Public Signal 0:019 Zy 1 Past Demand Level 0:043 At 1 Past Productivity 0 "Z y Annual Demand Shock 0:274 Residual 0:305 *in absolute values 4 Conclusion This paper addresses the issue of agricultural commodity price changes throughout the past decades. In particularly, the paper takes the case of corn for US markets. It derives two analyses: long run and short run. In the …rst, it initially runs a sign restricted VAR as a benchmark, …nding that demand and speculation are the biggest drivers of annual changes. It then contrasts this with a simple model in which it simulates theoretical prices, such that demand and supply match each year. That is, it estimates the price that would have theoretically cleared the market (no inventory changes), then compares it against the observed prices. The intention was to develop what one could consider a proxy for a non-distorted price and compare to see whether the observed price was too far away from it. Results show that after the 1996 farm bill, the relationship between these two time series is very close. That is, there does not seem to be a major di¤erence between the market- clearing price and the observed price. Before 1996, observed prices clearly di¤ered from theoretical ones. There are several explanations for this, but a major one is that government agriculture policy 37 intervened heavily, mostly through price-sustaining policies. These were instrumented by either government accumulation of stocks (the CCC program) or subsidizing the private sector to do that for them (the FOR program). After 1996, government policy became more market friendly by introducing price ‡ s budget available for oors below equilibrium and limiting the government’ stocks. Interestingly, annual changes are mostly explained by supply shocks, in contrast ith the VAR approach. When looking at overall changes through time, Figure II.A shows that no big distortions are to blame for price hikes in the 2000s. Moreover, a more detailed analysis shows that the emergence of an energy-related source of demand for corn had a signi…cant impact, and explains an important aspect of price increases in 2005-2010. This is in line with the increasing positive correlation between corn prices and energy ones. An important thing to note is that government mandates could have played a major role in this. Therefore, it could be that we are observing a shift in regulation type from direct purchases to private coercion. Identifying these interventions, quantifying them, and estimating their economic impact are subject’s for future research. The second part of the paper analyzes the short run. It develops a quarterly based model in which agents can only produce at certain quarters, as in most agricultural markets. It then linearizes it and analyzes second order moments. When focusing on a higher frequency, there is space for speculation to play an important role, since it is possible to make pro…ts by storing for short periods of time and reselling. In other words, it is irrational for inventories to go up forever, but it is perfectly possible that they increase for one or two quarters. Therefore, the paper proceeds by modeling speculation and introduces a theoretical innovation: information shocks. Given that storage is by de…nition related to expectations about future prices, the information available at each moment is key to inventory decision-making. The model has two sources of information: public and private. Public information comes as a forecast by a government agency (in this case, the USDA). Private information comes from information markets (consulting …rms, private research departments, etc.). These reports help agents form expectations about what future prices might be and make decisions accordingly. Given that USDA reports are publicly available, and that they contain quantitative forecasts for demand and supply, the model is able to estimate their impact on market prices. It also identi…es the precision of private signals and estimate a time series for them. Results show that as one would expect, demand and supply shocks account for a large fraction of price changes— as much as 52%. However, this leaves a relevant space for speculation or inventory 38 shocks. Information shocks (which act by inducing agents to buy or sell their stored goods) can explain an additional 18%. A remaining 31% is due to factors not included in the model but that alter storage decisions, as well as second-order e¤ects in demand or supply. The second section quanti…es by how much these unexplained factors can in‡uence observed prices. Finally, another important result is that non-explained sources of price changes, whatever these may be, do not seem to be any more relevant in the 2000s than they were before. This again con…rms once more that the idea that nonfundamental factors have increased in importance is not sustainable. Results show that in the short run, prices can deviate severely from fundamentals, even though in the long run that is not the case. Given that the market analyzed here is well developed in terms of liquidity and transparency, this raises questions regarding how informative prices may be in the short run, with important implications with respect to GDP accounting and theoretical modeling. As in Romer (2016), my results challenge the notion of market prices as a result of a standard pro…t- and utility-maximizing framework. In conclusion, the evidence presented here indicates that there is no evidence for the hypothesis that corn markets arti…cially deviated from fundamentals in the long run, but does allow the possibility for quarterly frequencies. As an innovation, this paper also shows the important role information shocks play in the short run. Unknown sources of variability are quanti…ed, but are yet to be explained. Further research in this respect is necessary. 39 5 References Acharya, V.V., Lochster, L.A. (2013), Limits to arbitrage and hedging : Evidence from commodity markets. Journal of Financial Economics, vol. 109, issue 2, 441-465. Basak, S. & Pavlova, A. (2016), A Model of Financialization of Commodities, Journal of Finance, Vol. LXXI, No 4. Baumister C. & Hamilton J.D. (2019), Structural Interpretation of Vector Autorregressions with Incomplete Information: Revisiting the Role of Oil Supply and Demand Shocks.American Economic Review, 109 (5): 1873-1910. Caldara, D. ,Cavallo M. & Iacoviello M. (2016), Oil Price Elasticities and Oil Price Fluctuations, International Finance Discussion Papers 1173, Board of Governors of the Federal Reserve System (U.S.). Fattouh, B., Kilian L., and Mahadeva L. (2013), The Role of Speculation in Oil Markets: What Have We Learned So Far?, The Energy Journal, vol. 34, issue 3. Bohl, M.T. & Stephan, P.M. (2012), Does Futures Speculation Destabilize Spot Prices? New Evidence for Commodity Markets, The Journal of Future Markets, August, 696-714. Bosch, D. & Pradkhan, E. (2015), The impact of speculation on precious metals futures markets, Resources Policy, v 44, 118-134. Chari, V.V. & Christiano, L.J. (2017), Financialization in Commodity Markets, Federal Reserve Bank of Minneapolis, Sta¤ Report 552. Chahrour, R. & Kyle, J. (2018), News or Noise? The Missing Link, American Economic Review, 108 (7): 1702-36. Egelkraut, T.M., Garcia, P. , Iriwn, S.H. and Good, D.L. (2003), An evaluation of Crop Forecast Accuracy for Corn and Soybeans: USDA and Private Information Agents, Journal of Agricultural and Applied Economics, April 35,1:79-95. 40 Hansen, L.P. & Sargent, T.J. (1993), Seasonality and approximation errors in rational expectations models, Journal of Econometrics, 55-21. Hussman, J.P. (1992), Market E¢ ciency and Ine¢ ciency in Rational Expectations Equilibria: Dynamic E¤ects of Heterogeneous Information and Noise, Journal of Economic Dynamics and Control, 16: 655-680. Iqbal, Z. & Babcock, B.A. (2018), Global Growing Area Elasticities of Key Agricultural Commodities Estimated Using Dynamic Heterogeneous Panel Methods, Agricultural Economics, September, Vol 49, Issue 5, 547-668. Kilian, L. & Murphy, D. (2014), The Role of Inventories and Speculative Trading in the Global Market for Crude Oil, Journal of Applied Econometrics, April, 29(7753) Knittel, C.R. & Pyndick, R. (2016), The Simple Economics of Commodity Price Speculation, American Economic Journal: Macroeconomics , 8(2): 85–110. Mallory, M. L., Irwin, S. H., & Hayes, D. J. (2012). How market e¢ ciency and the theory of storage link corn and ethanol markets. Energy Economics, 34(6), 2157-2166. Mehra, R. (2012), Consumption-Based Asset Pricing Models, The Annual Review of Financial, Economics 4:385–409. Milgrom, P. & Stokey, N. (1982), Information, Trade and Common Knowledge, Journal of Economic Theory, vol. 26, issue 1, 17-27. Pindyck, R. (1990), Inventories and the Short-Run Dynamics of Commodity Prices, The RAND Journal of Economics,vol. 25, no. 1, 1994, pp. 141–159. Roberts, M. J., & Tran, A. N. (2012). Commodity Price Adjustment in a Competitive Storage Model with an Application to the US Biofuel Policies (No. 323-2016-11729). Roberts, M.J., and Wolfram, S.(2013). "Identifying Supply and Demand Elasticities of Agricultural Commodities: Implications for the US Ethanol Mandate. " American Economic Review", 103 (6): 2265-95. 41 Romer, D. (2006), Do Firms Maximize? Evidence from Professional Football, Journal of Political Economy, 114(2):340-365. Scheinkman, J. & Xiong, W. (2003), Heterogenous Beliefs, Speculation and Trading in Financial Markets, Paris-Princeton Lectures in Mathematical Finance 2003, Lecture Notes in Mathematics, vol 1847, pp 217-250 Sockin, M. & Xiong, W, (2015), Informational Frictions and Commodity Markets, The Journal of Finance, October, Vol. LXX, No. 5. Tegene, A., Hu¤man, W. E., & Miranowski, J. A. (1988). Dynamic corn supply functions: A model with explicit optimization. American Journal of Agricultural Economics, 70(1), 103-111. Townsend, R.M. (1983), Forecasting the Forecast of Others, Journal of Political Economy, Vol. 91, No 4, pp. 546-588. Zhou, W., & Babcock, B. A. (2017). Using the competitive storage model to estimate the impact of ethanol and fueling investment on corn prices. Energy Economics, 62, 195-203. 42 6 Appendix 6.1 Additional Figures Figure III Figure III. Fig. III shows the contribution of each factor in total price change (measured in percentage points change with respect to the previous year). More speci…cally, Fig. III shows changes in prices due i ; A ,! and to productivity shocks. They were calculated by replacing the estimated values of Zt , i 2 t i ffood,feed,ethanol,exportsg ; into equation (2). Each term in equation (2) is represented by a di¤erent bar for a given year. 43 Figure V Figure V. Fig. V shows the contribution of total price change due to private inventory purchases. It was calculated by estimating the change in prices not explained by equation (2), i.e., the residual between explained price changes and observed price changes. That residual was then multiplied by the proportion of corn inventories not held by the government under the CCC program. 44 Figure VI Price Indexes for di¤erent Commodities 45 Figure VII Figures VI-VII. Fig. VI shows how corn and soybean prices have deviated from other commodities that were not used for energy production. Fig. VII shows increasing daily correlation between corn prices and oil prices for a past-…ve-years rolling window. 46 Figure XII Figure XII. Fig. XII shows the e¤ect on prices for forecasts regarding supply. The e¤ects are shown as changes in prices in percentage points. 47 Figure XIII Figure XIII. Fig. XIII shows the e¤ect on prices for forecasts regarding demand. The e¤ects are shown as changes in prices in percentage points. 6.2 High-Frequency Model.Proof of parameter estimates Parameters that associate state variables with prices were calculated using the method of unde- termined coe¢ cients. Below are the steps and formal results of such calculations for each quarter. The solving pattern is identical in all four quarters; therefore, I mainly describe the …rst quarter and solve the following ones in the same way. Since …nal expressions are implicit, solutions were found numerically. 6.2.1 First quarter Expectations equation in quarter 1 results: 48 h i E1 (p2 ) = II E ~1 + A II E "Z II E [X ] II E A II E Z II E [Z ] : 1 1 2 1 2 + 3 1 2 + 4 1 2 + 5 1 2 + 8 1 y Replacing the speculator’s policy function in the previous equation: h i p1 II E ~1 + II E A Z 2X2 + = 1 1 A 2 1 "Z 2 + II E [X ] 3 1 2 + II E 4 1 2 + II E 5 1 2 + II E [Z ] : 8 1 y h i p1 II E ~ II E A Z = 1 1 A1 + 2 1 "Z 2 + II 3 2 X2 + II E 4 1 2 + II E 5 1 2 + II E [Z ] : 8 1 y I now replace X2 with the market-clearing equation for the …rst quarter and leave the price variable on the left-hand side so to match the linear solution equation: 2 h i 3 II E A ~1 + II Z 6 2 E1 " 2 3 7 1 1 6 2 7 6 ~1 p [Zy + "z ] + X1 7 6 6 p 1 A 1 1 7 7 p1 =66 + 3 II 2 4 5+ 7 7: 6 1 +( Ap1 + Z1 p1 0 1 )p1 7 6 7 4 5 II A II Z II + 4 E1 2 + 5 E1 2 + 8 E1 [Zy ] 2 3 II A 1 1 + A A0 6 h i 7 6 7 h i 6 + II 2 ~1 p1 A p1 [Zy + "z + 7 1 II 1 0p 1 6 3 1] + X1 7 p1 3 2 ( Ap 1 + Z1 1 ) =6 7: 6 II E A II E Z 7 6 + 4 1 2 + 5 1 2 7 4 5 II Z + 8 z Zy 1 + 1 2 3 II A + II II Z 6 1 1 1 A A0 + 8 1 + 8 z Zy 1 7 6 7 h i 6 II 2 p1 Zy + + II 2 p1 A1 7 ~ 1 II 1 0p 1 6 3 3 7 p1 3 2 ( Ap 1 + Z1 1 ) =6 7: 6 II 7 6 3 2 p1 " z 1+ II 3 2 X1 + 7 4 5 II E A II E Z + 4 1 2 + 5 1 2 2 3 II 2 ~1 p1 A II 2 p1 " z 6 3 3 1 7 6 7 h i 6 + II 2 X1 7 1 II 1 0p 1 6 3 7 p1 3 2 ( Ap 1 + Z1 1 ) =6 7: 6 II A II Z II 7 6 + 1 1 + 2 1 + 1 A A0 7 4 5 + II Z II 2 p1 Zy 8 z y 1 3 Therefore, the …nal price solution equation is: 49 II 2 p1 II 2 p1 3 ~1 3 p1 = h iA h i "z 1 1 II 1 0p 1 1 II 1 0p 1 3 2 ( Ap 1 + Z1 1 ) 3 2 ( Ap1 + Z1 1 ) II 2 II 3 1 A +h i X1 + h i 1 1 II 1 0p 1 1 II 1 0p 1 3 2 ( Ap 1 + Z1 1 ) 3 2 ( Ap1 + Z1 1 ) II II 2 Z 8 z +h i 1 +h i Zy 1 1 II 1 0p 1 1 II 1 0p 1 3 2 ( Ap 1 + Z1 1 ) 3 2 ( Ap 1 + Z1 1 ) II II 2 p1 1 A 3 +h i A0 h i Zy : 1 II 1 0p 1 1 II 1 0p 1 3 2 ( Ap 1 + Z1 1 ) 3 2 ( Ap 1 + Z1 1 ) And parameter values are given by: II I h ( 3 2)p1 i: 1 = 1 1 0p 1 ( 3 2)( Ap1 + Z1 1 ) II I h ( 3 2 )p1 i: 2 = 1 1 1 ( II 3 2)( Ap1 0p + Z1 1 ) I h ( II 3 2 ) i: 3 = 1 ( II 3 2)( Ap1 1 + Z p1 1 ) II I = h 1 i: 4 1 1 1 ( II 3 2)( Ap1 0p + Z1 1 ) II I = h 2 i: 5 1 1 1 ( II 3 2)( Ap1 0p + Z1 1 ) II I = h 8 z i: 6 1 1 1 ( II 3 2)( Ap1 0p + Z1 1 )) II I = h 1 A i: 7 1 1 1 ( II 3 2)( Ap1 0p + Z1 1 ) II I h ( 3 2 )p1 i: 8 = 1 1 1 ( II 3 2)( Ap1 0p + Z1 1 ) 6.2.2 Second quarter The expectations equation is: E2 (p3 ) = ~1 III A + III E "Z III E [X ] III E A III E Z III E [Z ] : 1 2 2 3 + 3 2 3 + 4 2 3 + 5 2 3 + 8 2 y 50 The solution equation and the speculator’s policy function are: II p2 ~1 III A III X III A III Z III Z : 2X 3 + = 1 + 3 3 + 4 2 + 5 2 + 8 y p2 ~1 III A III II III A III Z III Z : = 1 + 3 2 X3 + 4 2 + 5 2 + 8 y I introduce the market-clearing equation and solve: h i p2 ~1 III A III II ~2 + Z p 1 III A III Z III Z : = 1 + 3 2 p2 Z 2 p2 + X 2 + 4 2 + 5 2 + 8 y 2 3 ~1 III A III Zy + "2 II 2 p2 h i 6 1 3 z 7 6 7 p2 1 Z2 p2 1 III 3 II 2 =6 6 + III3 II 2 X 2 + 7: 7 4 5 A III Z + III Z + III 4 2 + 5 2 8 y 2 3 ~1 III A III II 2 p2 " 2 h i 6 1 3 z 7 6 7 p2 1 Z2 p2 1 III 3 II 2 =6 6 + III 3 II 2 X2 + III A + 4 2 7: 7 4 5 + III Z + III III II 2 p2 Zy 5 2 8 3 Final price solution equation for quarter 2 is: 2 III 3 h III 1 i A1 h ( 3 2)p2 i "2 6 1 1 1 1 z 7 6 Z2 p2 ( III 3 2) Z2 p2 ( III 3 2) 7 6 ( III 2) III 7 p2 = 6 h 3 i X2 + +h 1 i A 7 6 + 1 4 Z2 p2 1 ( III II 2) Z2 p2 1 ( III 2) 2 7 6 3 3 7 4 III ( III ( III 2)p2 ) 5 +h 1 5 i Z + h 8 3 i Zy : 1 II 2 1 1 Z2 p2 ( III 3 2) Z2 p2 ( III 3 2) Parameter solutions are: III II = h 1 i: 1 1 1 Z2 p2 ( III 3 2) III II h ( 3 2)p2 i: 2 = 1 1 Z2 p2 ( III 3 2) III II h ( 3 2) i: 3 = 1 1 II Z2 p2 ( III 3 2) 51 III II = h 4 i: 4 1 1 Z2 p2 ( III 3 2) III II = h 5 i: 5 1 1 II Z2 p2 (III 3 2) III III II ( h 8 ( 3 2)p2 )i: 8 = 1 1 Z2 p2 ( III 3 2) Third quarter is identical to the second one 6.2.3 Fourth Quarter Solution equation and expectations equation are respectively: p ~4 = ~1 IV A + ~4 IV Z + IV X 4 + IV A 4 + IV Z 4 + IV Z : y 1 2 3 4 5 8 2 h i 3 IE ~1B + A I E ["z ] + I E [X ] 6 1 4 2 4 1B 3 4 1B 7 6 7 p1B ] = 6 E4 [~ 6 + IE 4 4 A 1B + IE 5 4 Z 1B + I 6 E4 [Zy 7 1] 7 : 4 h i 5 + IE ~1 + + A I E [Z ] 7 4 8 4 yB Introducing the speculator’ s policy function into the expectations equation I get the following: 2 h i 3 I ~ I z I 6 1 E4 A1B + 2 E4 ["1B ] + 3 E4 [X1B ] 7 | {z } 6 | {z } 7 6 =0 7 6 ~1 +E4 " A A 7 6 A | {z }1 B 7 6 7 6 A 7 6 4 7 6 I A I Z ~ 7 II 2X 1 B + = 6 p4 6 +| 4 E4 1B {z + 5 E4 1B + 6 Zy + 7 A1 + 7 I I 7: 6 } 7 6 =0 7 6 7 6 + I E [Z ] 7 6 8 4 | {z } yB 7 6 7 6 Z + E [" z ] 7 4 z y 4 | {z1B} 5 Z 4 II p4 I I ~1 + I A I I I Z IX : 2X 1 B + = 1 A + 7 A 1 4 + 6 + 8 z Zy + 8 4 + 3 1B Now, I proceed to solve by introducing the market-clearing equation: 52 p4 I I ~1 + I A I I I Z I II = 1 A + 7 A 1 4 + 6 + 8 z Zy + 8 4 + 3 2 X 1B : 2 3 I + I ~1 + A I A + I + I Zy p4 6 1 A 7 1 4 6 8 z 7 =4 5 0 1 + I Z + I II 2 ~4 + X4 + Z p p4 Z pt : 8 4 3 4 4 2 3 I + I ~1 + A I A + I + I Zy + I Z p4 6 1 A 7 1 4 6 8 z 8 4 7 =4 5 I II 0 1 + 3 2 p4 (Zy + "z 4) + X4 + Z4 p4 pt : 2 3 I + I ~1 + A I + I Zy + I A + I Z p4 6 1 A 7 6 8 z 1 4 8 4 7 =4 5: I II 0 1 + 3 2 p4 (Zy + "z 4 ) + X4 + Z4 p4 pt 2 3 I + I ~1 + A I + I Zy + I A + I Z 6 1 A 7 6 8 z 1 4 8 4 7 6 7 p4 =6 6 I 3 II 2 p4 Zy 3 II 2 p4 " z 4+ 7: 7 4 5 II II 0 1 + 3 2 X4 + 3 2 Z4 p4 pt 2 h i 3 I + I ~1 + A I + I I II 2 p4 Zy 7 6 1 A 7 6 8 z 3 6 7 pt = 6 7: p4 I II 0 1 2 Z4 p4 I II 3 6 3 2 p4 " z 4+ 7 4 5 I II I A I Z + 3 2 X4 + 1 4 + 8 4 I I II 1 A + 7 ~1 3 2 p4 p4 = h iA h i "z 4 1 I II 0 1 1 I II 0 1 3 2 Z4 p4 3 2 Z4 p4 I II I 3 2 1 A +h i X4 + h i 4 1 I II 0 1 1 I II 0 1 3 2 Z4 p4 3 2 Z4 p4 h i I I I II I 6 + 8 z 3 2 p4 8 Z +h i 4 + h i Zy : 1 I II 0 1 1 I II 0 1 3 2 Z4 p4 3 2 Z4 p4 I I IV h 1 A+ 7 i: 1 = 1 II 0 1 ( I 3 ) 2 Z4 p4 I II IV h ( 3 2)p4 i: 2 = 1 II 0 1 ( I 3 2) Z4 p4 53 I II IV h [ 3 2 ] i: 3 = 1 II 0 1 ( I 3 ) 2 Z4 p4 I IV = h 1 i 4 1 II 0 1 ( I 3 2) Z4 p4 I IV = h 8 i 5 1 II 0 1 ( I 3 2) Z4 p4 I+ I II IV [h 6 8 z ( 3 2)p4 ] i 8 = 1 II 0 1 (I 3 2) Z4 p4 54