Commodity Co-movement, Producer Diversification, and Output Volatility

Transcription

1 Commodity Co-movement, Producer Diversification, and Output Volatility Nicolas Merener & Eugenia Steglich Universidad Torcuato Di Tella Presentation at UWO - September 2015 Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 1 / 21

2 Commodity correlations Introduction Figure: Commodity return correlations, from Tang and Xiong 2012 Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 2 / 21

3 The question Introduction Large increase in commodity volatility and commodity correlations in last decade. 90% of Russia s output is oil. Brazil s is 50% oil, 18% soybeans, 7 % corn, sugar... Focus on volatility of national output value. How does risk differs across commodity producers? How has it changed? Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 3 / 21

4 Intuition from finance Introduction A well diversified portfolio is less risky. Output value for well diversified producer should be less volatile. An increase in volatilities should harm every producer. An increase in correlations should harm diversified producers more. Economic magnitude of this effect? Simple mapping from concentration to risk? Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 4 / 21

5 Approach Introduction Consider 21 globally traded commodities. Identify 50 countries with largest commodity production / GDP ratio. Construct time series of commodity output value for each country. Perform empirical analysis of output volatility and its relationship with diversification. Compare with simple decomposition of output volatility in terms of diversification and overall market conditions. Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 5 / 21

6 Related literature Introduction di Giovanni and Levchenko (2009, RESTAT) (2010, NBER). Impact of trade on manufacturing output volatility. Exports as risky portfolio. Malik and Temple (2009, J. Dev. Econ.) Geography of output volatility, Koren and Tenreyro (2007, QJE) on volatility and development. This paper: the interaction between concentrated production, commodity dynamics and commodity output dynamics. Borensztein et al. (2013) on gains associated with hedging against commodity price volatility at the national level. Merener (2015, J. Futs. Mkts) Higher concentration of production increases commodity jump risk. Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 6 / 21

7 Commodity data Commodity dynamics Production data by national origin and with yearly frequency between Price data for , monthly. Commodity Production data Price data Volatility Aluminum USGS LME Lead USGS LME Nickel USGS LME Copper USGS LME Zinc USGS LME Tin USGS LME Oil BGS NYMEX Platinum USGS NYMEX Palladium USDA NYMEX Soybeans USDA CBOT Wheat USDA CBOT Corn USDA CBOT Coffee USDA NYBOT Cotton USDA NYBOT Cocoa FAO NYBOT Sugar USDA NYBOT Orange juice USDA NYBOT Palm oil FAO BME Rubber FAO TCE Canola USDA ICE Rough Rice FAO CBOT Average Average by World Prod Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 7 / 21

8 Commodity dynamics (Partial) commodity correlation matrices Alum. Copper Oil Plat. Soy Corn Coffee Cotton Avg. Copper Oil Platinum Soy Corn Coffee Cotton Rice Alum. Copper Oil Plat. Soy Corn Coffee Cotton Avg. Copper Oil Platinum Soy Corn Coffee Cotton Rice Alum. Copper Oil Plat. Soy Corn Coffee Cotton Avg. Copper Oil Platinum Soy Corn Coffee Cotton Rice Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 8 / 21

9 Output variance Basket return decomposition A country produces N commodities, with sizes Q i (t), i = 1,...N Commodity market prices p i (t), monthly. The value of national output is Monthly return is V (t) = Q i (t)p i (t) i=1 r(t) = i=1 Q i(t + 1)p i (t + 1) i=1 Q i(t)p i (t) i=1 Q i(t)p i (t) Production measures are updated yearly. Decompose basket return on changes in prices and changes in quantities (last term only in January) r(t) = i=1 Q i(t)(p i (t + 1) p i (t)) i=1 Q i(t)p i (t) + i=1 p i(t)(q i (t + 1) Q i (t)) i=1 Q i(t)p i (t)... and sane decomposition if we look at yearly returns Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 9 / 21

10 Basket return and weights Output variance Define w i = Q i p i (t) j=1 Q jp j (t) with i=1 w i = 1. Basket return due to price changes is r(t) = w i r i (t) where r i (t) is the return of the i th commodity. The basket return variance σ 2 B is written in terms of the variances and correlations of individual commodity returns i=1 σ 2 B = i=1 w 2 i σ 2 i + i=1,j i w i w j σ i σ j ρ ij Our goal is to characterize the variance of the basket return in terms of overall market conditions and the degree of concentration implicit in {w 1,..., w N }. Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 10 / 21

11 Measure of concentration Output variance Adopt Herfindahl index (H) as measure of concentration H = i=1 w 2 i, A large H indicates a highly concentrated commodity producer. Effective number of commodities produced by a country N Effective = 1 H, If national output equally distributed among N commodities, w i = 1/N, then N Effective = N. Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 11 / 21

12 Output variance Concentration and basket return variance σ 2 B = i=1 w 2 i σ 2 i + i,j=1,j i w i w j σ i σ j ρ ij Many commodities, too many parameters. Take cross sectional averages. i=1 σ w i World i=1 w i World σ i 1 ρ N (N 1) i,j=1,j i ρ ij Approximate decomposition of basket variance in terms of the structure of production H, average commodity volatility and average correlation (overall market condition) {σ 2, ρ}. Similar to di Giovanni and Levchenko (2009). σ 2 B σ 2 N i=1 = σ 2 N i=1 w 2 i w 2 i + σ 2 ρ = σ 2 H + σ 2 ρ(1 H). i,j=1,j i w=1 w i w j, + σ 2 ρ(( w i ) 2 w=1 w 2 i ), Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 12 / 21

13 Output variance Concentration and basket return variance σb 2 σ 2 ρ + Hσ 2 (1 ρ). Non-linear on time-varying market conditions {σ, ρ} and production structure H. Linear in H for fixed {σ, ρ}. Slope and constant term should vary in time as {σ, ρ} change. Figure: Approximate basket variance and concentration. Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 13 / 21

14 Preliminary calculations Results Based on yearly production data and average yearly prices select countries with commodity production to GDP ratio above 5% during Compute yearly Herfindahl index for each country using production data. For each of the 50 countries in the sample compute the price and quantity components of monthly and yearly basket returns i=1 Q i(t)(p i (t + 1) p i (t)) i=1 Q i(t)p i (t) i=1 p i(t)(q i (t + 1) Q i (t)) i=1 Q i(t)p i (t) where Q i (t) is real production of commodity i in previous year. Keep yearly production fixed in the computation of the 12 returns within a year. Break sample in three periods: , , Compute basket return statistics (variance, etc) for each country and period. Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 14 / 21

15 Results (Partial) list of countries per Basket/GDP % Commodity / GDP HH HH HH Avg HH N eff Argentina 11.0% Bangladesh 8.7% Bolivia 11.1% Brazil 7.5% Canada 7.8% Chile 11.6% China 15.2% Ecuador 23.2% Egypt 23.7% Ethiopia 8.9% Ghana 15.5% India 9.2% Indonesia 31.2% Iran 32.7% Kuwait 59.4% Malaysia 54.7% Mexico 13.4% Nigeria 44.7% Norway 15.6% Pakistan 9.0% Russia 24.0% Saudi Arabia 58.5% Ukraine 6.9% Venezuela 37.2% Avg. correlation between Commodity / GDP and Herfindahl: 61.1 % Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 15 / 21

16 Results Concentration and sources of volatility Figure: Sources of output value volatility and concentration Cross-sectional averages of output value volatility: Total return annual vol: 34.6%. Fixed price annual vol: 12.8%. Fixed quantity annual vol: 30.9% Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 16 / 21

17 Results Concentration and basket variance Figure: Basket variance due to price changes and concentration per period Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 17 / 21

18 Results Cross-sectional output variance and concentration Three time periods: , , Sample: 50 commodity producers with highest commodity production to GDP ratio. Basket variance computed on monthly basket returns keeping fixed quantities. Motivated by σ 2 B σ 2 ρ + Hσ 2 (1 ρ), run robust OLS over the the cross section of commodity producers for CommodityProd GDP BasketVariance k = β 1Concentration k +β 2 +β 3 +β 4OilDummy k +Const+ɛ k GDP k WorldGDP k Real Theo Real Theo Real Theo. Concentration (0.019) (0.014) (0.011) CommodityProd GDP (0.021) (0.012) (0.012) GDP WorldGDP (0.64) (0.15) (0.04) Oil producer (0.009) (0.007) (0.006) Constant (0.007) (0.005) (0.004) Sample size R-squared Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 18 / 21

19 Results Commodities in a basket and volatility Figure: Effective number of commodities and volatility, Figure: Effective number of commodities and volatility, Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 19 / 21

20 Results Commodities in a basket and volatility Average volatility for commodity producers with low and high effective numbers. Effective number of commodities Mean basket volatility Standard error (0.014) (0.010) (0.007) Number of countries Effective number of commodities 3 Mean basket volatility Standard error (0.008) (0.008) (0.005) Number of countries Average Commodity Output / GDP ratio for our sample is 19% in first two periods, 21% in third period. Increase in output price volatility leads to increase in GDP volatility. Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 20 / 21

21 Conclusions Conclusions We explore the risk faced by commodity producing nations. The bulk of output value volatility is due to price fluctuations rather than output fluctuations Concentrated producers face higher output value volatility than diversified producers (10% gap in annual volatility, historically). However, the recent increase in commodity correlations had a significantly negative effect on diversified producers (6% vol. increase) and a milder effect in concentrated producers. The mean gap in risk across producers has narrowed to 6% in annual volatility. An approximate decomposition of output variance sheds light in the incidence of commodity volatility, correlations and the degree of concentration. Nicolas Merener & Eugenia Steglich (UTDT) Commodities and Output Volatility 21 / 21