Financialization and Commodity Markets 1

Transcription

1 Financialization and Commodity Markets 1 V. V. Chari, University of Minnesota Lawrence J. Christiano, Northwestern University 1 Research supported by Global Markets Institute at Goldman Sachs.

2 Commodity prices Observations since 2000, trend and volatility appear to have changed. Figure: log(producer Price Index: All Commodities/PCEPI) Monthly Data Trade in commodity futures markets. since 2000, volume of trade has increased substantially.

3 Question What is the empirical link between financialization and the behavior of commodity prices? Time series at best only suggestive because it consists of one observation. Cross-sectional evidence may be more informative.

4 Empirical Method Use information on a cross-section of commodities. Construct and study a panel dataset with 131 commodities over 20 years. Huge variation in futures markets across commodities Many commodities not traded at all in futures markets. Among traded commodities, much variation in trade volume. Advantage of studying cross-section: can potentially distinguish between Was the change in price behavior a consequence of the increased volume in futures markets? Or, was it a consequence of other factors ( growth in China? ) that affected all commodities?

5 Answers What is the empirical link between changes in spot price behavior and changes in volume of trade? No systematic association. Do traded (i.e, more financialized) commodities exhibit higher price volatility than non-traded commodities? Traded commodities exhibit modestly less volatility. But, literature shows that trade volume does matter for futures returns (Hong-Yogo). How can trade volume matter for futures prices but not spot prices? Can theory account for all these observations? Yes. A classic theory can account for these observations.

6 Measuring Financialization Notation for futures markets: S L S s H L H s : number of long positions (e.g., bushels of wheat ) held by non-commercial traders ( outsiders ) : number of short positions of outsiders : number of long positions held by commercial traders ( insiders ) : number of short positions held by insiders Data from CFTC on all trades in organized futures exchanges in the United States Would like to have data on over-the-counter and overseas markets.

7 Measuring Financialization Two indicators of financialization: Open interest: S L + H L (= S s + H s ) Net financial flows: S L S s ( = H s H L). Each indicator scaled by world production of relevant commodity. Futures trades Bakers Net flows Wheat farmers Wall Street

8 Index Construction Construct aggregate index of prices and financialization. Each commodity price is divided by GDP deflator. P t = 131 i=1 w i P it, P i,t : i th commodity price/gdp deflator w i = 1 T t ( P it Y i,t j P jt Y j,t Also compute aggregate indices of financialization in the same way: ). oi t = 28 S L i,t w + HL 28 i,t S L i,t i, nff t = Y i=1 i,t w Ss i,t i, w Y i = w i i=1 i,t 28 : world production of commodity i in year t Y i,t j=1 w j

9 CFTC Data Sources Volume of trade on all commodity futures contracts on organized exchanges in the US. For each CFTC commodity, we identify measure of world production Some issues: HOGS in CFTC matched with pig crop, PORK BELLIES with pig crop. Indices of World Production and Prices. Fuels: British Petroleum website. Minerals: US Geological Survey. Food and softs: Food and Agriculture Organization of United Nations (FAOSTAT) Some cases, do not have price indices for world production, so used US price.

10 Commodity Price Index Behavior Our commodity price index behaves similarly to the BLS s Producer Price Index: log, real spot price Figure 1: Broad Indicators of Commodity Prices Producer Price Index: All Commodities (left scale) Our constructed commodity index (right scale) log, real spot price

11 Financialization Behavior Indices of open interest and net flows open interest jumped from on average one-half of world production to 2.5 times world production. net financial flows rose only a tiny bit. 3 Figure 2: Indices of Commodity Trade Volume ratio Open Interest Net Financial Flows

12 Source of Increase in Open Interest Most of the higher volume is increased intra-group trade within outsiders and within insiders. S L oi = 0.27 HL oi = 0.73 Outsiders share of open interest is growing, but it s small S L oi

13 Motivation for Analyzing Individual Commodities Consider some statistic of prices (e.g., volatility), y i, and some statistic of financialization (e.g., open interest), x i, in 2 periods. Suppose y i = β x i + d + ε i, where d is a common shock (e.g., d ~ Growth in China ). Consider aggregate relation: y = β x + d + ε, y = 1 n n i=1 y i, etc. Problem: cannot uncover β with only one observation. Data at level of individual commodities potentially informative. Regression of y i on x i : if ε i x i, slope is β and d absorbed in constant term.

14 Outline Impact of financialization on: Dynamics of spot prices. Dynamics of futures returns. Interpret the results in the light of a model.

15 Analysis of Commodity Spot Price Dynamics and Financialization Two approaches. Structural Break approach. Centered Moving Average approach.

16 Comparisons of Means and Standard Deviations Mean of mean log P in post- and pre- 2000s: X and Ȳ X Ȳ, s X Ȳ = s s2 2. T 1 T 2 s X Ȳ Used large N, small T asymptotic theory from Ibragimov and Müller, (2010, 2011), t r, where r is degrees of freedom adjustment that allows for possible change in standard deviation across samples. Also, used a bootstrap procedure to compute p values. Analogous large N, small T inference for difference in variance across samples.

17 Table 1: Mean and Volatility Properties of Commodity Price Growth Commodity Group n Mean, logarithmic growth Variance, log growth di (p-value: theory, boot) di (p-value: theory, boot) All (21, 22) ( 6, 12) Indexed (25, 24) ( 22, 29) Not-Indexed (18, 19) ( 1, 2) Traded (21, 22) ( 3, 6) Non-traded (24, 23) ( 8, 17) Note: (i) n denotes the number of commoditie in the specified group. (ii) p-value indicates the probability, under the null distribution of no di erence, of getting an even higher di erence than was realized in the data. (iii) p-values are reported using a particular sampling theory and by a bootstrap procedure for robustness (see the appendix for details). 1

18 Table 1: Mean and Volatility Properties of Commodity Price Growth Commodity Group n V di (p-value: theory, boot) di (p-value: theory, boot) All (21, 22) ( 6, 12) Indexed (25, 24) ( 22, 29) Not-Indexed (18, 19) ( 1, 2) Traded (21, 22) ( 3, 6) Non-traded (24, 23) ( 8, 17) Note: (i) n denotes the number of commoditie in the specified group. (ii) p-value indicates the probability, under the null distribution of no di erence, of getting an even higher di erence than was realized in the data. (iii) p-values are reported using a particular sampling theory and by a bootstrap procedure for robustness (see the appendix for details). 1

19 Table 1: Mean and Volatility Properties of Commodity Price Growth Commodity Group n V di (p-value: theory, boot) di (p-value: theory, boot) All (21, 22) ( 6, 12) Indexed (25, 24) ( 22, 29) Not-Indexed (18, 19) ( 1, 2) Traded (21, 22) ( 3, 6) Non-traded (24, 23) ( 8, 17) Note: (i) n denotes the number of commoditie in the specified group. (ii) p-value indicates the probability, under the null distribution of no di erence, of getting an even higher di erence than was realized in the data. (iii) p-values are reported using a particular sampling theory and by a bootstrap procedure for robustness (see the appendix for details). 1

20 Table 1: Mean and Volatility Properties of Commodity Price Growth Commodity Group n Mean, logarithmic growth Variance, log growth di (p-value: theory, boot) di (p-value: theory, boot) All (21, 22) ( 6, 12) Indexed (25, 24) ( 22, 29) Not-Indexed (18, 19) ( 1, 2) Traded (21, 22) ( 3, 6) Non-traded (24, 23) ( 8, 17) Note: (i) n denotes the number of commoditie in the specified group. (ii) p-value indicates the probability, under the null distribution of no di erence, of getting an even higher di erence than was realized in the data. (iii) p-values are reported using a particular sampling theory and by a bootstrap procedure for robustness (see the appendix for details). 1 Interestingly, traded goods are less volatile (though possibly not signifcantly so) than non-traded goods, suggesting that financialization stabilizes.

21 Table 1: Mean and Volatility Properties of Commodity Price Growth Commodity Group n V di (p-value: theory, boot) di (p-value: theory, boot) All (21, 22) ( 6, 12) Indexed (25, 24) ( 22, 29) Not-Indexed (18, 19) ( 1, 2) Traded (21, 22) ( 3, 6) Non-traded (24, 23) ( 8, 17) Note: (i) n denotes the number of commoditie in the specified group. (ii) p-value indicates the probability, under the null distribution of no di erence, of getting an even higher di erence than was realized in the data. (iii) p-values are reported using a particular sampling theory and by a bootstrap procedure for robustness (see the appendix for details). 1 This relatively aggregated data does not send a resounding signal. We move on now, to the fully disaggregated data.

22 Structural Break Approach: Regressions For each commodity, regress log real spot price on time trend with a break in Calculate the change in the slope coeffi cient. the standard deviation of the regression residual. Also calculate change in variance of commodity price growth. Relate above to change in: open interest net financial flows.

23 Change, spot price trend Change, spot price trend Change, spot price trend Change, spot price trend Figure 4b: Change in Trend and Net Financial Flows All commodities, slope = Change, Net Financial Flows Traded commodities, slope = Change, Net Financial Flows 0.2 Softs, slope = Minerals and fuel, slope = Change, Net Financial Flows Change, Net Financial Flows

24 Table 2: Change in Commodity Inflation Dynamics, 1990s to 2000s, as a Function of Change in Financialization P-value on when financialization measured with nff (P value with oi) change in commodity inflation dynamics t change in financialization u t variables in analysis change in variance of residual from time trend change in slope coefficient on time trend change in variance all commodities 64 (66) 11 (15) 39 (48) indexed 76 (89) 14 (50) 39 (47) non-indexed 20 (18) 62 (14) 17 (18) traded 72 (78) 12 (21) 44 (56) softs 12 (32) 2 ( 4) 2 ( 9) minerals and fuels 76 (68) 24 (32) 67 (68) Notes: (i) two measures of financialization - net financial flows (nff) and open interest (oi). (ii) p-value is the probability, under the null distribution that 0, of getting a value of higher than its empirical realized value. For details, see the appendix.

25 Table 2: Change in Commodity Inflation Dynamics, 1990s to 2000s, as a Function of Change in Financialization P-value on when financialization measured with nff (P value with oi) change in commodity inflation dynamics t change in financialization u t variables in analysis change in variance of residual from time trend change in slope coefficient on time trend change in variance all commodities 64 (66) 11 (15) 39 (48) indexed 76 (89) 14 (50) 39 (47) non-indexed 20 (18) 62 (14) 17 (18) traded 72 (78) 12 (21) 44 (56) softs 12 (32) 2 ( 4) 2 ( 9) minerals and fuels 76 (68) 24 (32) 67 (68) Notes: (i) two measures of financialization - net financial flows (nff) and open interest (oi). (ii) p-value is the probability, under the null distribution that 0, of getting a value of higher than its empirical realized value. For details, see the appendix. Finding : (except for softs: corn, lumber, etc.) there is no significant relationship between a structual break in price dynamics and change in financialization.

26 Centered Moving Average Approach Potential pitfall for structural break approach: it may be sensitive to the (somewhat arbitrary) choice of 2000 as the break date. Our second ( Centered Moving Average ) approach. Compute a rolling standard deviation of the growth rate of commodity prices (5-point moving average). A shortcoming of this approach is you lose some data.

27 Centered Moving Average approach Regress volatility time series on financialization measures. Done only for commodities for which there is non-zero volume of trade in each time period.

28 Figure 8: Response of Volatility to Two Measures of Volume (results based on individual traded commodities) mean response of volatility to Net Financial Flows = 0.24 mean response of volatility to Open Interest = Net Financial Flows Open Interest

29 Centered Moving Average approach Message of previous slide distribution of response of volatility to financialization is dispersed. centered on negative numbers: more financialization results in less volatility. mean standard deviation over all variables is 0.2. mean slope on open interest is raise open interest from one times world production to two, then volatility falls small amount, from 0.2 to 0.14.

30 Centered Moving Average approach The following figure displays scatter plot of all data in each category: all, softs, minerals&fuels. Here, we included non-traded commodities. For example, in the all category, we have observations.

31 Figure 9: Spot Price Volatility and Volume σ( log price) Figure 9a: All commodities, slope = σ( log price) Figure 9a: All commodities, slope = σ( log price) σ( log price) σ( log price) Figure 9b: Traded Net commodities, Financial Flows slope = Figure 9c: Soft Net commodities, Financial Flows slope = Figure 9c: Minerals Net Financial and Fuels, Flows slope = Net Financial Flows σ( log price) σ( log price) σ( log price) Figure 9b: Traded commodities, Open interest slope = Figure 9c: Soft commodities, Open interest slope = Figure 9c: Minerals Open and Fuels, interest slope = Open interest

32 Centered Moving Average approach Message of previous slide No evidence that increased financialization raises volatility. Indeed, the evidence suggests volatility may drop. But, these effects are quantitatively small here.

33 Table 3: Regression, Volatility of Commodity Prices on Intensity of Financialization volatility t intensity t intensity measure net financial flows open interest group of variables (95% conf interval) (95% conf interval) all commodities (-0.024,0.029) (-0.004,0.007) traded (-0.019,0.033) (-0.002,0.011) softs (-0.028,0.027) (-0.003,0.008) minerals and fuels (-0.060,0.108) (-0.014,0.028) Notes: (i) standard deviation based on centered, 7 point moving average of commodity price growth; (ii) data combines all observations on the group of commodities listed in left column; (iii) we dropped silver and gold from the analysis underlying this table. See the text for discussion. (iv) bootstrap confidence intervals described in text.

34 Table 4: Another Way to See that Financialization Has Little Impact on Spot Price Volatility (1) (2) Measure of financialization Measure of spot price dynamics 12 month average oi growth centered, 6 month moving average standard deviatio 2 nd quartile interquartile range associated with column (1) quarti lower bound mean (median) (-0.343) (6.400) upper bound rd quartile lower bound mean (median) (1.831) (6.876) upper bound

35 Summary So Far There has been increased financialization in commodity markets. Cross sectional analysis shows little evidence of a systematic correlation between the degree of financialization and properties of commodity prices. We now look at financialization and the dynamics of futures returns. Some have reported evidence that financialization has a substantial effect on futures returns.

36 Correlations Between Return on Futures and 3 Month Tbills/Equity Study impact of financialization on volatility of commodity futures returns and their comovement with other asset returns. Computed correlations of futures returns and other returns for each year. Regress time series of correlations on financialization measures for each commodity. Finding: mean coeffi cients nearly zero.

37 Table 4: Futures Returns and Financialization Correlation of Futures Returns with: Volatility of Financialization Equity Returns 3 month Tbill Futures Return measure month day month day month day nff (-4.49, 3.40) 0.26 (-2.44, 2.70) (-2.61, 2.51) (-0.33, 0.49) 0.21 (-0.22, 0.36) 0.01 (-0.06, 0.06) oi 0.34 (-0.45, 0.84) 0.12 (-0.33, 0.63) 0.12 (-0.44, 0.48) 0.01 (-0.09, 0.07) 0.01 (-0.05, 0.06) 0.01 (-0.01, 0.02) Notes: (i) numbers in parentheses are boundaries of 95 percent confidence interval compute under the null hypothesis that behavior of futures returns is unrelated to degree of financialization. Confidence interval computed using a bootstrap procedure described in the text. (ii) entries are averages across 27 commodities, of slope in regression of column variable on financialization measure. Regression for each of the 27 commodities was computed using annual observations, (iii) month (day) means that, for each year s observation the correlation or standard deviation of the return on a futures contract is based on monthly (daily) data for that year. (iv) oi - open interest, nff - net financial flows;

38 Relation to Tang and Xiong Tang-Xiong computed pairwise correlations between returns on commodity futures. We compute pairwise correlations by centered moving average, j = lag,..., lag, lag = 130 days in daily correlations, lag = 6 months in monthly data. Tang-Xiong found that the pairwise correlations were greater for indexed commodities and for non-indexed commodities. Concluded that financialization matters. We obtain similar findings for daily data, but differences between indexed and non-indexed commodities appear to go away in monthly data.

39 Next Slide, Pairwise Correlations in Daily Returns

40 0.4 average return correlation among commodities all commodities indexed commodities off-indexed commodities

41 Next Slide, Pairwise Correlations in Monthly Returns

42 0.6 average return correlation among commodities 0.5 all commodities indexed commodities off-indexed commodities

43 Financialization and Futures Markets Hong and Yogo (JFE, 2012) demonstrate a link between financialization and futures prices. Open interest helps to predict futures returns. Net financial flows do not help to predict futures returns. We have a different way to demonstrate these links.

44 Measuring the Importance of Financialization in Futures Markets Fictitious investor adopts following strategy: in month t, examine the volume of trade in commodity futures up to month t 1. go long in a basket of commodities that show the most volume of trade (hot strategy). two measures of volume of trade : net financial flows - net commercial trader shorts, divided by open interest. growth of open interest over the past year. Compare hot net financial flow strategy; hot open interest growth strategy; random strategy, random basket.

45 Cumulative returns from 3 futures contract strategies NOTE : Shaded areas represent 90 percent confidence interval 250 Mode, random futures market strategy Hot net financial flow strategy Hot open interest growth strategy

46 Finding: Consistent with Hong-Yogo, open interest growth contains substantial information about subsequent commodity returns.

47 Summary of Empirical Evidence There is substantial variation in the degree of financialization in the cross section of commodities. there does not seem to be a systematic relationship between the degree of financialization and spot price volatility. But, financialization does seem to matter for futures returns. The volume of open interest growth appears to reliably predict high subsequent futures returns. Net financial flows unrelated to subsequent futures returns. How could open interest affect futures returns without having a systematic effect on spot prices?

48 One Period Model Futures markets allow agents to reduce risk by hedging. Insiders: Farmers worry wheat prices, P, will be low. Bakers worry that P will be high. Outsiders: care about P because correlated with their own income. Futures market: opens when wheat planted, with price F. Futures return - P F. All market participants also speculate Maximize mean-variance utility subject to constraints. Solution: demand for long (short, if negative) contracts = hedging demand + speculative demand {}}{ E (P F) αvar (P F) α risk aversion, same for everyone

49 Commodity Market Bakers buy wheat, bake bread Outsiders No direct par<cipa<on in produc<on or use of commodity Farmers Plant seeds, grow wheat

50 Bakers Commodity Market Source of uncertainty for insiders: Demand for bread Θ + ε Realized at the end Known at beginning Outsiders Farmers

51 Bakers Commodity Market Source of uncertainty for insiders: Demand for bread Θ + ε Realized at the end Known at beginning Outsiders Farmers Source of uncertainty for outsiders: outside income, x, correlated with ε.

52 Commodity Market Bakers Hedging mo<ve: Uncertainty in price of wheat, P. Want to buy long in futures market. Hedging need limited because bread price is a natural hedge. Outsiders Farmers

53 Commodity Market Bakers Hedging mo<ve: Uncertainty in price of wheat, P. Want to buy long in futures market. Hedging need limited because bread price is a natural hedge. Outsiders Farmers Hedging mo<ve: Uncertainty in P. Want to sell short in futures market.

54 Commodity Market Bakers Hedging mo<ve: Uncertainty in price of wheat, P. Want to buy long in futures market. Hedging need limited because bread price is a natural hedge. Outsiders Farmers Hedging mo<ve: Hedging mo<ve: Uncertainty in P. Want to sell short in futures market. Cov(income,P) < 0, go long. Cov(income,P) > 0, go short. Covariance changes over <me.

55 Commodity Market Bakers buy wheat, bake bread Net financial Flows, n ff : Net purchases of long contracts by outsiders. Outsiders No direct par<cipa<on in produc<on or use of commodity Farmers Plant seeds, grow wheat

56 Commodity Market Bakers buy wheat, bake bread Farmers Plant seeds, grow wheat Net financial Flows, n ff : Net purchases of long contracts by outsiders. Futures return, P F Outsiders No direct par<cipa<on in produc<on or use of commodity Data: Cov (P-F, n ff ) = 0 Shocks to outsider hedging demand: cov < 0 Shock to insider hedging demand: cov > 0.

57 Lack of Pattern Across Markets Between Financialization and Spot Price Volatility There is a panel of markets in the cross section. Each market has a different mix of insider and outsider hedging demand shocks. Exogenous variations in measure of outsiders. If most hedging demand shocks are to insiders, then outsiders stabilize price volatility. Otherwise, outsiders destabilize. Endogenize outsider participation decision. Relation between participation and spot price volatility ambiguous. Example: small increase in shock volatility from a point where most shocks are to hedging demand by insiders. direct effect (e.g., holding measure of outsiders constant): raise price volatility. entry effect: by increasing outsider participation, stabilize price volatility.

58 Conclusion Little evidence of a systematic relation, across commodity markets, between financialization and volatility. Some modest evidence that volatility is lower with financialization. Financialization has a systematic impact on commodity markets Gross volume of futures market trades predict futures returns (though, net financial flows do not). We described a simple (fairly standard) model to interpret the observations.

59 Formal Statement of the Model and Results First, a formal statement of the problems of the agents. Then, formal statement of the model implications for: covariance, open interest and futures market returns. covariance, net financial flows and futures market returns lack of systematic relationship across markets between financialization and spot price volatility.

60 Agents in the Model Measure one of farmers produce wheat. Measure one of bakers produce bread from wheat. Measure µ of outsiders participate in futures markets for wheat.

61 Timing in the Model Beginning of Period: Anticipated component of demand for bread, θ, realized. Anticipated component of outsiders income, s, realized. then: End of period: Farmers choose how much wheat to produce. Futures market meets. Unanticipated shocks, η and ν, realized. Unanticipated component of demand: ε η + ν. Unanticipated component of outsiders income: η. All shocks independent of each other.

62 Demand for bread Demand and Technology P Q = D (Q, θ + ε), where P Q price of bread, Q quantity of bread. Technology for producing bread from wheat, q : Q = q δ, 0 < δ < 1. Cost function for producing wheat: c (q) = cq cq2, c, c > 0. All agents have mean-variance preferences over consumption, z : Ez α var [z] 2

63 Farmer s Problem Conditional on θ and s, farmers choose q and H w to solve max E [Pq + RH w ] α 2 var [Pq + RHw ] c (q), H w quantity of wheat futures bought. P spot price of wheat. R return on long futures contract, Solution: R = P F. F = c (q) H w = hedging demand {}}{ q + speculative demand {}}{ ER. αvar (R)

64 Baker s Problem End of period problem: first order condition: max P Q q δ Pq. q δp Q q δ 1 = P. profits at optimum: ( ) 1 δ 1 Pq. Beginning of period problem: [ ( ) ] 1 max E Pq H b δ 1 + RH b α2 [ ( ) ] 1 var Pq δ 1 + RH b. solution: H b = hedging demand { ( }}{ q 1 1 ) δ + speculative demand {}}{ ER. αvar (R)

65 Outsider s Problem Outsiders income, x, given by x = sη. outsiders income is partially correlated with unanticipated component of demand for bread, ε = η + v. correlation varies with realization of s. leads to fluctuation in outsiders hedging demand. Outsiders futures market problem, conditional on θ and s : solution: max H o E [RH o sη] α 2 var [RHo sη]. H o = hedging demand {}}{ s σ2 η σ 2 ε speculative demand {}}{ ER +. αvar (R) Note: if σ 2 η = 0, then outsiders have no hedging demand.

66 Futures Market Price Determination Market clearing condition: H w + H b + µh o = 0. Induced demand function for wheat using bakers fonc: ( ) P = D q δ, θ + ε δq δ 1. We work with linearized representation: P = D 0 D q q + θ + ε. Yields linear equilibrium solutions for equilibrium prices and quantities.

67 Equilibrium Solutions Futures return (R = P F) : Production decision: Lemma: R = R 0 + R θ θ + R s s + ε, q = q 0 + q θ θ + q s s. R θ > 0, R s < 0 and q θ, q s > 0. Open interest, oi, given by: oi = 1 [ H w + H b ] + µ H o. 2 Net financial flows, nff, given by: nff = µh o.

68 Open Interest and Net Financial Flows Proposition: there exists σ 2 s such that cov (nff, R) = 0. Basic idea of proof: s shocks drive R(= P F) down and nff up θ shocks drive R up and nff up. Proposition: Suppose H b, H o > 0 for all realizations of shocks and µ not too large, then cov (oi, R) > cov (nff, R) Basic idea of proof: When θ goes up, farmers go short and, since µ small, bakers must go long, H b up, R up. When s goes up, outsiders go long (so, R down), insiders go short (H b down).

69 Var(P) and Exogenous Participation Variance of spot prices, var (P), given by: δc + ασ2 2 ε 2+µ δ ( ) σ 2θ c + + δd q D q + ασ 2 ε 2+µ ( µ ) ασ 2 η δ ( c + D q ) + ασ 2 ε 2+µ 2 σ 2 s + σ 2 ε. Proposition: If σ 2 θ small, outsiders destabilize spot prices. recall, F = c (q). F and so q respond more to s, so P more variable. Proposition: If σ 2 η or σ 2 s small, outsiders stabilize spot prices. F and so q respond more to θ, so P more stable. Increase in outsiders may stabilize or destabilize. Depends on market details.

70 Endogenizing Outsider Participation Outsiders have fixed cost, k, of entering each futures market. Enter before any shocks realized. Enter if surplus, S = U P U np, from participation exceeds cost: S k. Let k denote fixed cost of marginal entrant. Equilibrium condition: F (S) = µ.

71 Endogenizing Outsider Participation Holding µ fixed, S is increasing in σ 2 θ and σ2 s. If s large and µ small, S is decreasing in σ 2 ε. Decompose overall effect on var (P) into direct effect and entry effect. Direct effect: holds µ fixed, varies parameter. Entry effect: holds parameter fixed, varies µ. Overall effect ambiguous if direct and entry effect have opposite sign.

72 Endogenizing Outsider Participation Proposition: if σ 2 η small, increase in σ 2 θ or σ2 s has ambiguous effect on var (P). Proposition: if σ 2 θ small, s large and µ small, increase in σ2 ε has ambiguous effect var (P). Our model consistent with absence of systematic relationship across markets between nff and var (P).

73 Literature and Our Contribution Substantial disagreement in the literature Financialization does not matter: Hamilton and Wu (2012), Irwin and Sanders (2011), Killian and Murphy (2013). Financialization does matter: Buyuksahin and Robe (2013), Hong and Yogo (2012), Tang and Xiong (2012). Evidence of market segmentation: Acharya et al (2011) and Etula (2010) Literature only looks at traded commodities. typically uses only a small subset of traded commodities. each paper uses different measures of financialization and different methods for measuring its impact. Our contribution: construct a panel dataset of prices, quantities and measures of financialization 131 traded and non-traded commodities over 20 years use variation over time series and cross section to investigate importance of financialization.