An Analysis of Illiquidity in Commodity Markets

Transcription

1 An Analysis of Illiquidity in Commodity Markets Sungjun Cho, Chanaka N. Ganepola, Ian Garrett Abstract We examine the liquidity premium demanded by hedgers and the insurance premium demanded by speculators. Our results indicate that both hedgers and speculators demand a higher premium in illiquid commodities, to provide insurance and liquidity, respectively. Using the Brennan, Huh & Subrahmanyam (2013) decomposition of illiquidity into turnover and size components, we find evidence suggesting there may as well be a size premium associated with the long-term insurance premium, such that speculators demand a larger insurance premium for smaller commodities. We also find that the liquidity premium demanded by hedgers for illiquid commodities varies across bullish and bearish markets with hedgers demanding a larger premium from speculators trading in illiquid commodities in bearish markets. Keywords: Hedging, insurance, speculators, hedgers, illiquidity premium JEL classification: Q02, G13, Sungjun Cho and Ian Garrett are affiliated with Alliance Manchester Business School, the University of Manchester. Chanaka Ganepola is affiliated with Alliance Manchester Business School, the University of Manchester and the Central Bank of Sri Lanka. This research however, do not necessarily reflect views of the Central Bank of Sri Lanka. Authors would like to thank participants at 2018 Commodity Markets Winter Workshop. Correspondence: Alliance Manchester Business School, Booth Street East, Manchester, M13 9SS, UK; sungjun.cho@manchester.ac.uk, chanaka.ganepola@manchester.ac.uk, ian.garrett@manchester.ac.uk 1

2 1 Introduction Research in commodity markets tends to emphasize the importance of hedgers and speculators, who allow these markets to function smoothly. In any market there are participants who consider themselves as market makers as they provide liquidity to the market and market takers, who consume liquidity. Chang (1985), Hirshleifer (1990) and Basu & Miffre (2013) among others, argue that commodity hedgers (consisting of producers and consumers of commodities) 1 demand liquidity in order to eliminate price risk of holding commodities. Speculators provide liquidity by taking positions as per hedgers demand. In the process, hedgers transfer their price risk to speculators and speculators earn a premium as reward. Therefore speculators are considered liquidity providers to the commodities market. However, Kang, Rouwenhorst & Tang (2017) find that the provision of insurance and liquidity do not refer to the same risk premium. Rather, they argue that there are two independent risk premia in commodity markets, one that rewards the investor for providing liquidity and one that rewards the investor for providing insurance. They suggest that speculators provide insurance to hedgers while hedgers provide liquidity to speculators. In particular, they argue that the change in net long positions explains liquidity provision from hedgers to speculators while the smoothed hedging pressure 2 (net short futures position) explains insurance provision from speculators to hedgers. Kang et al. (2017) argue that hedgers demand a higher premium when speculators request short-term liquidity in relatively illiquid commodities. However, Kang et al. (2017) do not examine whether speculators demand a premium from hedgers when they wish to hedge position in commodities that are illiquid. From the theory of normal backwardation we know that speculators absorb the price risk of hedgers at a certain cost. The asset pricing literature finds that the illiquid assets require a higher expected return than liquid ones. One contribution of our paper is we investigate whether the same applies in commodity markets, that is if hedgers seek protection in illiquid commodities, they should pay a larger premium compared to liquid commodities, in order to reward the additional risk taken by the speculators. We find that speculators demand 1 Producers, merchants, processors and users who use futures for hedging purposes are known as commercial traders by the US Commodity Futures Trading Commission (CFTC). Non-commercial investors include speculative traders. disaggregatedcotexplanatorynot.pdf 2 It follows from the theory of normal backwardation that, demand for price insurance by hedgers are reflected in their net short futures position. Conversely, speculators are net long when futures prices are expected to increase, thereby earning the premium as compensation for the risk absorbed from hedgers. 2

3 a premium to absorb the price risk only for illiquid commodities. Given the importance of liquidity to our analysis, a second contribution we make is to examine the sensitivity of our results to the measure of liquidity. A concern with the seminal Amihud (2002) measure of illiquidity is that it can be distorted by market capitalization (Cochrane 2005). This is a particular concern for our analysis given the significant growth in the total size of commodity markets over the period Florackis, Gregoriou & Kostakis (2011) and Brennan et al. (2013) propose a solution to this problem. Florackis et al. (2011) use stock turnover rather than dollar volume 3 when calculating illiquidity while Brennan et al. (2013) decompose the Amihud measure into two components: absolute returns earned per unit of turnover, which is a turnover version of the Amihud measure, and market capitalization. We make use of this decomposition to examine if there is both a liquidity effect related to turnover and a size effect in commodity markets and whether any such effects carry additional premia. Specifically, we test the hypothesis that the liquidity (insurance) premium paid by speculators (hedgers) differs depending on the degree of illiquidity of commodities. We also test the hypothesis that a larger premium is required to induce investors to take positions in small commodities as compared to large ones. We find that hedgers (speculators) require an additional premium for short-term liquidity (insurance) in illiquid commodities. We also find that speculators provide insurance to hedgers at a lower cost, in large commodities compared to smaller commodities. A considerable number of publications discuss the size effect in the stock market, including Banz (1981), Blume & Stambaugh (1983) and Fama & French (1993). These suggest that investments is small cap stocks are more rewarded compared to large cap stocks, implicating that small cap stocks are more riskier compared to the large cap stocks. We argue the same with regard to the commodity market, that there may be a large premium required for investors (speculators/hedgers) to take positions in small commodities compared to large ones. As explained earlier on the two return premiums found by Kang et al. (2017), the liquidity premium is the reward for fulfilling short-term liquidity needs of speculators and the insurance premium is for providing insurance to hedgers in the long-term. The size affect may therefore reflect in the short-term liquidity premium and/or the long lived insurance premium. Our results suggest that, the short-lived liquidity premium is indifferent across commodities with different sizes. However, the speculators may provide insurance to hedgers at a lower cost, in large commodities 3 Brennan et al. (2013) identifies this measure as the turnover version of the Amihud measure. 3

4 compared to smaller commodities. A final question we consider is whether there is a negative relationship between liquidity and returns, that is, whether the liquidity premium is higher in down markets rather than up markets. Hameed, Kang & Vishwanathan (2010) find that liquidity is negatively related to stock market returns. They find that liquidity is affected more when stock returns are negative than when they are positive. Brennan et al. (2013) compare the illiquidity of the stock market on days where returns are negative to those days where returns are positive. They find that illiquidity on down market days is significantly priced in stock returns, unlike illiquidity on up market days. One possible explanation for this is offered by Brunnermeier & Pedersen (2009), market financiers tighten the funding requirements during bear markets so that investors may have to sell their holdings even at a lower price, depending on the urgency of the margin requirement. Therefore, traders seeking liquidity in down markets pay a larger premium compared to up market days. We ask the question of whether speculators pay a larger premium for liquidity in down markets compared to up markets. If financiers tighten funding requirements as per Brunnermeier & Pedersen (2009), then it should also affect hedgers as they are another group of traders in the market. Therefore, hedgers should also pay a larger insurance premium during down markets, should the funding constraints apply to them. Our results show that only speculators incur a higher cost when buying illiquid commodities in bear (down) markets. The insurance cost of illiquid commodities for hedgers on the other hand, is not affected by whether markets are bullish or bearish. These findings are consistent with Daskalaki & Skiadopoulos (2016) who find that, increase in margins, which is the reaction of financiers in down markets according to Brunnermeier & Pedersen (2009), affect positions held by speculators rather than the positions held by hedgers. The remainder of the paper is organized as follows. Section 2 discusses the sources of data and construction of variables. Section 3 describes the empirical findings in detail. Section 4 concludes the paper. 2 Data We use the data provided in the commitment of traders (COT) report, published weekly on Friday by the CFTC. The report provides details on features the positions taken by commercial 4

5 investors (hedgers) 4 and non-commercial investors (speculators) for each commodity in our sample, at market close on every Tuesday. We extract weekly long, short and total open interest of each hedgers and speculators, for 24 commodities from 06/10/ /12/2015. We download the daily futures prices of nearest (first) and next-nearest (second) futures contracts for each commodity, from the Bloomberg database. The commodities in our sample are distributed over five categories. These are; agriculture( Wheat, Corn, Oats, Soybean, Soybean meal, Soybean oil, Rough rice); metal (Copper, Platinum, Palladium, Gold, Silver); softs (Coffee, Cotton, Cocoa, Sugar, Orange juice, Random length lumber); energy (Crude oil, Natural Gas, Heating oil); and livestock (Lean hogs, live cattle, feeder cattle). We also obtain the returns on the S&P500 index from CRSP database for the above dates. 2.1 Construction of Commodity Futures Returns We start by constructing a continuous series of excess returns for each commodity. We assume a rollover to the second futures contract from the first futures contract on the 7th calender day 5 of the maturity month in order to avoid the possibility of the futures prices being affected by expiry/delivery effects. Therefore, we hold the first futures contract until the 7th calender day of the maturity month. On the 7th calender day of the maturity month, we rollover to the second futures contract, which will then become the first futures contract. The futures excess return R i,t of commodity i at time t is given by, R i,t = F i(t, T ) F i (t 1, T ) F i (t 1, T ) (1) where, F i (t, T ) is the price of the futures contract of commodity i at time t, which matures at time T. We then compound the daily returns, from Tuesday of week t until Tuesday of week t + 1 in order to construct weekly returns. 2.2 Hedging Pressure and Net Change in Long Positions (Net Trading) Our research closely follows Kang et al. (2017) and therefore we follow them and construct the two variables that represent the premium paid by speculator to hedgers for liquidity provision 4 Recall from footnote 1 that, producers, merchants, processors and users who use futures for hedging purposes are known as commercial traders by US CFTC. Non-commercial investors include speculative traders. The literature also identifies commercial traders as hedgers and non-commercial traders as speculators. 5 We rollover on the next business day if 7th day of the maturity month is not a business day 5

6 and the premium paid by hedgers to speculators for providing insurance. Kang et al. (2017) estimate the hedging pressure HP i,t for commodity i at time t as: HP i,t = SP i,t H LPi,t H (2) OI i,t where, the superscript H represents hedgers, OI i,t is the open interest okang et al. (2017) then use the trailing 52-week moving average of HP i,t as the risk factor to capture the premium speculators require for providing insurance to hedgers. We call this smoothed hedging pressure (HP i,t ) Kang et al. (2017) define a second risk factor, the net trading measure (Q t ), that represents the premium required by hedgers for providing short-term liquidity to speculators; Q H i,t = netlp H i,t netlp H i,t 1 OI i,t 1 (3) Where, netlpi,t H = LPi,t H SPi,t H 2.3 Illiquidity Measures We initially use the Amihud measure Amihud (2002) as the measure of illiquidity in the commodities market. This measure is widely used in commodities literature (see Daskalaki, Kostakis & Skiadopoulos (2014) and Kang et al. (2017), for example.). The measure for a commodity i at time t is given by; Amihud i,t = r i,t DV OL i,t (4) Where, r i,t is the daily futures return of commodity i at time t and DV OL i,t is the dollar value of the traded contracts of commodity i at time t. We estimate the daily Amihud measure and take its weekly average as our weekly illiquidity measure for each commodity. Brennan et al. (2013) recommend the use of a turnover-based version of the Amihud measure (Amihud T ). The turnover-based Amihud measure comes from decomposing the original Amihud measure as follows (Brennan et al. 2013): 6

7 Amihud = r DV OL = r turnover turnover DV OL = r shares traded shares outstanding shares traded shares outstanding shares traded price per share = r turnover 1 size = Amihud T size 1 Brennan et al. (2013) originally proposed this decomposition for stocks. Therefore we are required to propose a commodity market futures equivalent for the number of shares outstanding in order to replicate this decomposition for commodity markets. We consider the open interest, which by definition is the total number of outstanding contracts available in the market and therefore serves as a measure of the commodity market equivalent of the number of shares outstanding. Therefore, the measures of turnover and size that we use for commodities are defined as: turnover = No.of traded futures contracts open interest (5) size = open interest contract size futures price (6) 2.4 Control Variables Following Kang et al. (2017), we use three control variables in our analysis of the presence and the behavior of the liquidity provision and insurance risk premia. They are: the basis, lagged futures excess returns (to control for any momentum effects) and the idiosyncratic risk of futures returns. Basis and momentum are considered as inventory-related risk factors by Gorton, Hayashi & Rouwenhorst (2013). Gorton et al. (2013) find that the commodity basis is negative when inventories are high and vice versa and secondly, they claim that positive 7

8 basis portfolios outperform negative basis portfolios in terms of returns 6. This implies that a negative shock to inventories may raise Expected futures return. This increase in the expected return is temporary as futures prices adjust upon the restoration of inventories. In the meantime, traders witness momentum in futures prices. However, this situation may change in a period where speculators are highly active in the market. When speculators expect futures price to increase following a drop in inventory level, they take positions in order to benefit from the price momentum 7. Therefore, excessive trading by a large number of speculators may restore inventories swiftly, compared to times of ordinary speculator activity. Quick restoration of inventories may eliminate basis trading and momentum trading opportunities. We construct the basis following Daskalaki et al. (2014). The basis of a commodity i at time t is; Basis i,t = F i(t, T 1 ) F i (t, T 2 ) F i (t, T 1 ) (7) where, F i (t, T 1 ) is the price of first futures contract and F i (t, T 2 ) is the price of second futures contract. T 1 and T 2 are the maturity dates of first and second futures contracts. Motivated by Kang et al. (2017) and Bessembinder (1992) 8, we also include a variable to proxy for unsystematic risk in commodity futures returns as a control variable. This variable is defined as S i,t V i,t where, V i,t is the volatility of the residuals from a 52-week rolling window regression of the futures return on commodity i on the return on the S&P500 index and, S i,t = 1 if speculators are net long and S i,t = 1 if speculators are net short in a commodity i at time t. Descriptive statistics for the variables are reported in table 1. Our estimates of mean and volatility of returns, the absolute net trading measure Q and hedging pressure HP are constructed following Kang et al. (2017) and therefore they are much similar to their summery statistics for the period 1994/01/ /11/01. However, we have expanded our dataset to 06/10/ /12/2015 period and therefore we find some differences with theirs. We find that on average positive returns are earned in all commodities except Rough rice, Wheat, Sugar, 6 Basu & Miffre (2013) also demonstrate that the basis, is an important factor in explaining futures excess returns. 7 Erb & Harvey (2006) find evidence on the presence of momentum in commodity market.moskowitz, Ooi & Pedersen (2012) and Dewally, Ederington & Fernando (2013) among others find that speculators are momentum traders. Further, Kang et al. (2017) explain that speculators in commodity market are momentum investors. 8 Hirshleifer (1988) argues that there is a hedging-pressure-dependent premium for unsystematic risk embedded in futures prices in addition to systematic risk premia. He further states that the sign of this premium depends on the net position held by the hedgers/speculators. The premium is negative if the hedgers(speculators) are net long(short) and vice versa. Bessembinder (1992) empirically finds evidence to support this. 8

9 Lumber, Natural gas and Lean hogs. The average net trading measure and hedging pressure appear to be high among metals, while Live cattle and Natural gas report the lowest (negative) hedging pressure. Both illiquidity measures, Amihud and Amihud T suggest that Oats is by far the most illiquid commodity among the 24 commodities in our sample. According to Amihud, the two most liquid commodities are Wheat and Feeder cattle. However, Amihud T suggest that the most liquid commodities are Gold and Crude oil. These two commodities are by far the largest commodities in terms of dollar value. Similarly, Amihud suggests that palladium is more illiquid compared to platinum. However, Amihud T, that measures illiquidity as the futures return per unit turnover, implies the opposite. This is mainly due to the fact that Platinum market is much larger compared to Palladium. The difference between the two measures lends some support to the idea that size affects the original Amihud measure. 3 Results Kang et al. (2017) claim that there are risk premiums associated with short-term liquidity provision and with insurance provision. To examine this, they estimate the following Fama & MacBeth (1973) cross-sectional regression and test the significance of β 1 and β 2. R i,t+1 = β 0 + β 1 Q H i,t + β 2 HP i,t + β 3 R i,t + β 4 Basis i,t + β 5 S i,t V i,t + ɛ i,t+1 (8) Table 2 reports the findings of the cross-sectional regression (8). First, columns (1) and (2) investigate whether net trading pressure (Q t ) and hedging pressure (HP t ) on their own are significant while column (3) replaces hedging pressure with smoothed hedging pressure (HP t ), following Kang et al. (2017). The results suggest that net trading pressure, which captures the short-term liquidity premium speculators pay to hedgers in return for them providing liquidity, is significant and positively related to returns while hedging pressure is insignificant. Smoothed hedging pressure, which captures the insurance premium hedgers pay to speculators, is significantly related to returns which raises the interesting question of whether net trading pressure and smoothed hedging pressure are jointly significant or whether one captures the effect of the other. Column (4) addresses this and shows that both are significant in explaining the cross section of excess returns on commodity futures. Thus, the evidence in table 2 suggests that both net trading pressure and smoothed hedging pressure earn a positive premium. 9

10 3.1 Illiquidity Effects on Commodity Markets In this section we examine to what extent illiquidity conditions matter in commodity markets. First, we use the estimated illiquidity proxy, ln(amihud) as the only independent variable (apart from the control variables) in the cross sectional regression (8). This result is reported in Column (1) of table 3 and it suggests that commodity-specific illiquidity is significantly priced in the cross section of futures returns. However, the significance of the Amihud factor disappears once the the two independent variables that represent the short term liquidity premium and the insurance premium are included in the regression (see column 2 in table 3.). This implies that the illiquidity return premium that we observed in column (1) is now absorbed by the short-term liquidity premium and/or the insurance premium. Therefore, we may expect to find an additional premium attached to the liquidity (insurance) premium demanded by hedgers (speculators) when they are to provide liquidity (insurance) to speculators (hedgers) for illiquid commodities. To investigate these possibilities, we follow Kang et al. (2017) and estimate two further regressions. The first regression is: R i,t+1 = β 0 + β 1 Q H i,t + β 2 Q H i,t D(Amihud) i,t + β 3 HP i,t + controls + ɛ i,t+1 (9) Where the dummy variable D(Amihud) i,t 9 takes the value 1 when commodity i is among the most illiquid ones at time t. This regression examines whether hedgers demand an additional premium for providing liquidity in particularly illiquid commodities. A positive and significant value for β 2 would suggest that they do. The results are reported in column (3) of table 3. These results are consistent with Kang et al. (2017) and suggest that the hedgers demand an additional premium to provide liquidity to speculators when they demand liquidity in the most illiquid commodities. According to the coefficients, the futures return is expected to increase by 3.6 basis points in the next week, if the net buying of hedgers in general increase by 1 percent of the total open interest. Moreover, this increase in futures return approximately doubles when hedgers increase long positions by 1 percent in illiquid commodities. To investigate whether speculators demand an additional premium from hedgers hedging particularly illiquid 9 Following Kang et al. (2017) we assign 1 to the dummy variable D(Amihud) i,t, if the trailing 52-week moving average of Amihud measure calculated for a particular commodity i is in the highest quartile among that of other commodities at time t, or zero otherwise. 10

11 commodities we repeat regression (9) for HP t : R i,t+1 = β 0 + β 1 Q H i,t + β 2 HP i,t + β 3 HP i,t D(Amihud) i,t + controls + ɛ i,t+1 (10) The results reported in column (4) of table 3 suggest that speculators demand a premium only when hedgers seek protection for illiquid commodities. Column (5) of table 3 reports results from estimating the regression with both interactive terms: R i,t+1 = β 0 + β 1 Q H i,t + β 2 Q H i,t D(Amihud) i,t + β 3 HP i,t +β 4 HP i,t D(Amihud) i,t + controls + ɛ i,t+1 (11) The estimated coefficients on Q H t and Q H t D(Amihud) do not change significantly from columns (3) while the estimated coefficient on HP t D(Amihud) is also unchanged from column (4) to (5). This implies that the return premiums: 1) received by hedgers for fulfilling short-term liquidity needs of speculators, 2) received by hedgers additionally for taking positions in illiquid commodities according to the demands of the speculators, and 3) received by speculators for providing insurance to hedgers in illiquid commodities, are statistically significant in explaining commodity futures returns. However, we find a possible inconsistency in the results reported in columns (4) and (5), with the hedging pressure hypothesis. The hedging pressure hypothesis suggests that speculators benefit from the premium paid by hedgers to absorb the future price risk of their commodity holdings. We find evidence consistent with this in table 2. Nonetheless, our findings in columns (4) and (5) of table 3 suggest that speculators may demand such a premium only if the commodity that hedgers seek protection is among the most illiquid commodities. A possible reason for this behavior is that, our proxy for illiquidity, the Amihud measure (Amihud 2002), could be masking the importance of HP t due to its size bias (Cochrane 2005). Moreover, it has been argued that, by construction, the Amihud measure computes the illiquidity per unit of market size (Brennan et al. 2013). The construction of our illiquidity dummy variable, D(Amihud), is based on the Amihud measure (Amihud 2002) given by equation (4). Brennan et al. (2013) show that Amihud s illiquidity measure for stock markets could be decomposed in to two separate components: 11

12 turnover and size (see the discussion in section 2.3.). This decomposition implies that large commodities may be considered as liquid even though its returns are high per unit turnover. Highlighting this issue, Brennan et al. (2013) show that it is more appropriate to consider the decomposition of the Amihud measure and use size as a separate variable, which allows one to account for illiquidity effects in the absence of potential size biases. Figure 1(a) plots the market-size-weighted ln(amihud) over time. The graph shows that market-size-weighted liquidity has increased post 2003 and that there are sharp fluctuations in market liquidity during Figure 1(b) plots the natural logarithm of total market size (ln(size)), over-time. We observe that the total market size was quite stable from 1993 to However, it appears that commodity markets as a whole expanded significantly in size between 2003 and Subsequently, the total market size decreased towards the end of 2009, perhaps due to a fall in the demand for commodities as a result of the global financial crisis. Further, the commodity market expanded during the next two years ( ) and maintained its size thereafter. The correlation between market size and the market-size-weighted ln(amihud) measure prior to 2003 is However, the correlation becomes 0.89 after This suggest that the Amihud illiquidity measure is substantially influenced by market size, especially after Therefore, the decrease in market size weighted ln(amihud) after 2003 could be mainly due to the expansion of highly liquid commodity markets. To further highlight this, figure 2, plots the average illiquidity, in terms of the Amihud measure (figure 2(a)) and the turnover version of the Amihud measure (figure 2(b)) over the sample period against the average size of each commodity. The dotted lines in figure 2 represent the 4th quartile of liquidity (horizontal) and size (vertical) and split each plot in to quadrants. Subsequently, quadrant 2 of each plot contains the most liquid, largest in terms of market size, while quadrant 4 contains the most illiquid, smallest commodities in terms of market size. Figure 2(a) (the traditional Amihud measure) shows that, there are no commodities in quadrant 1, which implies that all commodities that are large in size are more liquid compared to other commodities. However, commodities display qualities of illiquidity when their corresponding market sizes are relatively smaller, as we observe a number of commodities in both quadrants 3 and 4. Compare this with figure 2(b) (the turnover version of the Amihud measure.). In figure 2(b) we can see that Natural gas is considered to be an illiquid commodity despite its large market size. This suggests that even though there is a large market interest for Natural Gas, the actual number of trades are lower. The turnover-based Amihud measure also suggests that 12

13 Crude oil and Gold are the most liquid commodities in the sample. There are some changes to the contents of quadrants 3 and 4 from figure 2(a) to figure 2(b). The commodities, Rough rice, Cocoa and Palladium, considered to be illiquid under the usual Amihud illiquidity measure are not illiquid under the turnover-based Amihud measure of illiquidity. Instead, commodities such as Orange juice, Soybean oil, Soybean meal and Platinum are found to be illiquid on average. We also notice in figure 2 that the size of the largest commodities are substantially larger than other commodities markets. For instance, Crude oil and Gold, are the two largest markets in the sample on average, are almost twice as large as Natural gas, which is the third largest commodity market. We further examine this by estimating the the value weights allocated for the 6 largest commodities out of the 24 commodities in the sample. The 6 commodities that carries the highest weights in a given year are shown in figure 3. This figure suggests that in any given year from 1993 to 2015, the 6 largest commodities account for approximately 50% to 70% of the total size of the market. Soybean, Crude oil and Gold have been among the 6 largest commodities during the entire sample period. Soybean had the largest share of the total size of the commodities market during Gold dominated the commodities market until the year 2000, until the size of the Crude oil market exceeded the size of Gold. The weight on Crude oil in the entire commodities market increases during , which also coincides with the significant growth of commodity market size that we highlighted in figure 1(b). The above analysis suggests that the illiquidity that is being measured by Amihud is mainly driven by the size of the commodity market, especially when it comes to commodities that are among the largest. We further motivate the idea of decomposing the Amihud measure, by undertaking a portfolio based analysis. We double sort the sample of 24 commodities, first by size and then by the turnover version of the Amihud measure. The average returns on those portfolios and the corresponding average returns on high-minus-low illiquidity portfolios are reported in panel 1 of table 4. We observe a statistically significant difference between the returns on low and highly illiquid portfolios, in small commodity markets. We also construct double sorted portfolios by first sorting commodities according to illiquidity, and then based on market size. These results are reported in panel 2 of table 4. Results suggest that small-minus-big portfolio return is statistically significant for illiquid commodities. This implies that there is an illiquidity (size) premium associated with small (illiquid) commodities and therefore, illiquidity and size effects needs to be accounted for separately in the context of commodity markets. On the strength of the preceding evidence, we use the turnover version of the Amihud measure 13

14 (Amihud T ) to estimate illiquidity throughout the remainder of our study. We also analyze the market size effect on liquidity and insurance premia, separately. We reestimate regressions, (9), (10) and (11), replacing D(Amihud) i,t with D(Amihud T ) i,t. Results of these cross-sectional regressions are reported in columns (1), (2), and (3) of table 5. The explanation of results in column (1) is similar to our explanation in column (3) of table 3. That is, the hedgers demand an additional premium to provide short-term liquidity in highly illiquid commodities. Therefore, removal of the size effect from the Amihud measure does not appear to make any difference to the short-term premiums associated with liquidity provision. However, exclusion of the commodity market size from the Amihud illiquidity measure seems to have an effect on the insurance premium related to hedging pressure. We observe that in contrast to the coefficients on HP t in columns (4) and (5) of table 3, the same coefficients in columns (2) and (3) of table 5 are statistically significant. This implies that, there is a premium that hedgers pay speculators for insurance. Moreover, hedgers also pay an additional premium if they seek protection in illiquid commodities, which is almost as three times the insurance premium required for liquid commodities. These results are consistent with the findings of Liu & Yong (2005) for the options market. According to Liu & Yong (2005), traders require to go long (short) in more stock and borrow (lend) more money in order to replicate a call (put) option, which hedges their risk in illiquid stocks. This implies that there is a higher cost of insurance for illiquid stocks. The same could be said about the futures contracts on illiquid commodities. In this case we know that the speculators are traders who provide insurance to hedgers. Therefore, we conclude that speculators demand a larger premium to provide insurance to hedgers in illiquid commodities. We also observe a subtle increase in R 2 values from columns (3), (4) and (5) of table 3, to columns (1), (2) and (3) of table 5. This suggests that, the turnover version of the Amihud measure performs somewhat better compared to the original Amihud measure (Amihud 2002). We discuss the effects of commodity market size on the liquidity and insurance premiums separately in the following subsection. 3.2 The Size Effect Using double sorted portfolios we confirmed earlier that there is significant difference between returns on small and large illiquid commodities. Also, in the previous subsection, we observe indications of a possible association between the insurance premium and commodity market 14

15 size. A comparison of the regression coefficients on, Q H t and HP t in columns (4) and (5) of table 3, with (7) and (8) in table 5, shows that the insurance premium becomes statistically significant when the size factor is removed from the illiquidity measure. This suggests that size influences the coefficient of HP t more than that of Q H t. We test this empirically with the assistance of a dummy variable, D(size) which is 1 if the trailing 52-week moving average of the market size calculated for a particular commodity is in the highest quartile among that of other commodities, or zero otherwise, in the regressions given below. R i,t+1 = β 0 + β 1 Q H i,t + β 2 Q H i,t D(size) i,t + β 3 HP i,t + controls + ɛ i,t+1 (12) R i,t+1 = β 0 + β 1 Q H i,t + β 2 HP i,t + β 3 HP i,t D(size) i,t + controls + ɛ i,t+1 (13) Regression results are reported in columns (4) and (5) of table 5. Firstly, column (4) presents results for the hypothesis that the premium required by hedgers to provide liquidity to speculators is not dependent on the size of the corresponding commodity market. Results suggest that this hypothesis cannot be rejected. Secondly, we test the null hypothesis that there is no additional premium required by speculators to provide insurance to hedgers depending on the size of the corresponding commodity market results reported in column (5) show that the coefficient of HP t D(size) is negative and statistically significant at the 10% level which leads us to marginally reject the null hypothesis that the coefficient is zero. To examine the robustness of this result, we add Q H i,t D(size) i,t to regression (13). Results reported in column (6) show that the coefficient on HP t D(size) remains unchanged even after controlling for Q H i,t D(size) i,t. Thus, speculators do not charge a lower premium to provide insurance in large commodities. This suggests that size of the individual commodity markets only matters in the context of the insurance premium. The results also imply that the insurance premium demanded by speculators reduces significantly when hedgers seek protection in larger commodity markets. A net sale by hedgers of 1% of the open interest over the past year results in a 0.76% increase in returns over the following week 10. However, hedgers requiring protection in commodities 10 As pointed out in our earlier discussion, Kang et al. (2017) find that, lagged hedging pressure (HP t ) cannot explain futures returns. It is the smoothed hedging pressure (HP t ), the moving average of hedging pressure over the past 52-weeks that is capable of explaining futures returns 15

16 with large open interest would only pay a premium of 0.30% for every 1% increase in net sales by hedgers over the past year. This implies that the larger the size of the commodity market, the cheaper it is for hedgers to buy protection against future commodity price fluctuations. Consequently, we are able to form an argument in relation to the commodities market that is parallel to the findings of Banz (1981), Blume & Stambaugh (1983) and Fama & French (1993) in relation to stock markets. Commodities that are small in terms of their individual market size are riskier than the commodities with large markets, due to the lack of open interest. Therefore, speculators absorb a relatively higher risk when they are requested to provide protection in small commodities. As a result, the insurance premium that speculators demand is smaller in the case of large commodities. To investigate this further, we double sort commodities, first by size and then according to HP t. Average returns of the high and low HP t portfolios and their differences are reported in table 6. According to these results, in both small and large commodity groups, returns on high smoothed hedging pressure portfolios are significantly larger compared to returns on low smoothed hedging pressure portfolios. This results provides more support in favor of the hedging pressure hypothesis that, speculators are rewarded for accepting a higher level of price risk. We also observe that there is a statistically significant difference between the returns on small and large commodities only when the smoothed hedging pressure is high. This suggests that, the insurance premium for commodities, in which speculators are to absorb a relatively larger price risk, is also influenced by the size of the commodities market. That is, speculators demand a higher reward for commodities with higher price risk and the premium increases further if the market for the particular commodity is relatively small. One may also raise the issue as to why market size matters only in the case of insurance. An answer to this question may lie in the persistent nature of market size of each commodity. Table 7 exhibits the probability 11 of market size of each commodity being in the largest quartile among other commodities each year. Figures in table 7 suggest that commodities such as Soybean, Crude oil and Gold have been in the highest size quartile since We also observe that Corn is gradually making its way out of the highest quartile, while copper gradually makes its way in to the highest quartile. This suggests that market size is rather persistent and does not change substantially week-to-week. Therefore, market size is not very important in the context of short-term demand for liquidity, although it matters when it comes to demand for insurance, 11 This is the probability at which D(size) i,t = 1 of commodity i at time t 16

17 which is decided upon the net short-positions held by hedgers, over a longer period of time. As a final test in this section, we investigate which of illiquidity and market size prevails over the other in the context of premia associated with liquidity provision and insurance provision. We estimate the following regression, for which the results are reported in column (7) of table 5. R i,t+1 = β 0 + β 1 Q H i,t + β 2 Q H i,t D(Amihud T ) i,t +β 3 HP i,t + β 4 HP i,t D(Amihud T ) i,t Results indicate that coefficients on both Q H t + β 5 HP i,t D(size) i,t + controls + ɛ i,t+1 (14) and HP t, and their corresponding interactive dummy variables are statistically significant, while the size dummy variable that interacts with HP t is not significant. This implies that liquidity dominates the size factor. Therefore it is possible to conclude that, speculators are rewarded additionally for providing insurance in both illiquid and small (in terms of market size) commodities. However, the premium required by speculators for providing insurance in small markets is negligible compared to the additional insurance premium required by them in illiquid markets. 3.3 Illiquidity in Bullish and Bearish Markets An interesting argument in the literature is that the effects of illiquidity on stock returns during bull markets are significantly different from the effects of illiquidity during bear markets. Brunnermeier & Pedersen (2009) show that, in bearish markets intermediaries are more inclined towards increasing margin requirements and therefore influencing market liquidity. Hameed et al. (2010) find that illiquidity in the stock market, as represented by the bid-ask spread, increases when the market experiences losses. However, stock market illiquidity (wider bid-ask spreads) does not respond to positive market returns. These findings suggest that illiquidity has an asymmetric effect on stock returns and therefore motivates us to examine the asymmetric illiquidity effects on commodity market returns. We introduce two dummy variables, D(UP Amihud T ) i,t and D(DW Amihud T ) i,t where, D(UP Amihud T ) i,t is 1 if R i,t 0 and the trailing 52-week moving average of the turnover version of Amihud measure calculated for a particular commodity is in the highest quartile among that of other commodities, zero otherwise, and D(DW Amihud T ) is 1 if R i,t 0 17

18 and the trailing 52-week moving average of turnover version of Amihud measure calculated for a particular commodity is in the highest quartile among that of other commodities or zero otherwise. We estimate the two regressions below: R i,t+1 = β 0 + β 1 Q H i,t + β 2 Q H i,t D(UP Amihud T ) + β 3 Q H i,t D(DW Amihud T ) + β 4 HP i,t + controls + ɛ i,t+1 (15) R i,t+1 = β 0 + β 1 Q H i,t + β 2 HP i,t + +β 3 HP i,t D(UP Amihud T )+β 4 HP i,t D(DW Amihud T ) + controls + ɛ i,t+1 (16) The results from estimating (15) and (16) is reported in columns (1) and (3) of table Results in column (1) suggest that hedgers demand different premiums to provide liquidity in both bull (up) and bear (down) markets. In bull markets, hedgers do not charge an additional premium from speculators to provide liquidity. However, when markets are bearish, speculators incur an additional liquidity cost to trade illiquid commodity futures contracts. This illiquidity premium in bear markets is almost three times the liquidity premium (0.0302) in bull markets. Regression results in column (3) show that coefficients of HP i,t D(UP Amihud T ) and HP i,t D(DW Amihud T ) are statistically insignificant. This implies that speculators do not demand an additional premium for illiquid commodities in both up and down markets when it comes to insurance provisions. These findings are consistent with the findings of Brennan et al. (2013) for the equity market, that investors are more concerned about liquidity in bearish markets. We find that hedgers in particular, charge a larger premium when the prices of illiquid commodity futures contracts are decreasing. However, we do not observe a difference in insurance premium on illiquid commodities during bull and bear markets. In section 3.1 we found that, there is an additional premium required by speculators for providing insurance on illiquid commodities.therefore, we test the conjecture that, even though hedgers demand a larger premium for liquidity from speculators who seek trading opportunities in illiquid commodities, in bear markets, speculators demand a similar insurance premium for 12 In regression results presented in columns (2) and (4) of table 8, we control for market size in addition to the regression results in (1) and (3). However, the interpretation of the results are unchanged 18

19 illiquid commodities despite the market condition. We do this by estimating the following regressions: R i,t+1 = β 0 + β 1 Q H i,t + β 2 Q H i,t D(UP Amihud T ) + β 3 Q H i,t D(DW Amihud T ) + β 4 HP i,t + β 4 HP i,t D(Amihud T ) + controls + ɛ i,t+1 (17) Estimation results of the above regressions are reported in columns (5) of table 8. According to these results coefficients of Q H t D(UP Amihud T ) and Q H t D(UP Amihud T ) are statistically insignificant. However, the premium attached to purchase of insurance in illiquid commodities (β 4 ) is statistically significant. The results from estimating (15), as reported in column (1) of table 8, show that the coefficient on Q H t D(UP Amihud T ) is statistically insignificant (this is also the case when the same regression is estimated with market size controls as per column (2)). To examine the possibility that, Q H i,t D(UP Amihud T ) is masking the effects of Q H i,t D(DW Amihud T ), which was determined to have a greater impact on futures returns according to columns (1) and (2) of table 8, we estimate: R i,t+1 = β 0 + β 1 Q H i,t + β 2 Q H i,t D(DW Amihud T ) + β 3 HP i,t +β 4 HP i,t D(Amihud T ) + controls + ɛ i,t+1 (18) and test whether β 2 is zero. Results from estimating (18) are reported in column (6) of table 8. The results suggest that, speculators pay an additional premium (0.0742) for illiquid commodities in bear markets. Therefore, a speculator may incur a premium of ( ) in bear markets while taking positions in relatively illiquid commodities. The coefficient on HP t D(Amihud T ) also remains statistically significant. The above points to the conclusion that speculators holding positions in illiquid commodities, are substantially effected in bear markets due to their high cost of liquidity. Although there exists an additional cost to speculators for demanding positions in illiquid commodities in both bull and bear markets, the speculators additional cost in bear markets is significantly larger compared to the bull markets. However, despite the market condition, speculators do not demand an additional premium when they consider providing insurance to hedgers in illiquid commodities. Speculators charge an additional premium from hedgers for providing insurance 19

20 despite the bullish/bearish market condition. This also implies that short-term effects of illiquidity depend on previous market trends, whether it is bullish or bearish and, long-term effects are independent of the market trends. Possible explanations for these results can be found in Kang et al. (2017) and Brunnermeier & Pedersen (2009). Brunnermeier & Pedersen (2009) argue that initial losses (in down markets) to speculators may lead to liquidity dry-ups in the market either, due to investors withdrawing from markets in order to stop further losses or due to financiers increasing their borrowing margins, or both. The findings in Daskalaki & Skiadopoulos (2016) confirm that speculators reduce their positions following increased margins 13 and they do not react significantly to initial decreases. This is an indication of speculators closing out their positions following initial losses to avoid further losses due to tightened margins. As a result, speculators demand for liquidity increases. According to Kang et al. (2017), hedgers provide short-term liquidity to speculators at a premium and there is also an additional premium required by hedgers if they are to provide liquidity in relatively illiquid commodities. Therefore, when the demand for liquidity is much higher than before, hedgers may demand a much larger premium for illiquid commodities compared to up markets. In addition to this, the margin effect on the positions held by hedgers is almost negligible, compared to its effect on speculator positions (Daskalaki & Skiadopoulos 2016). This suggests that hedgers are not as alarmed by margin increases likely to have been caused by negative market returns. As a result, the cost of insurance to hedgers for illiquid commodities, does not differ across bear and bull markets, although they incur an additional insurance cost in general for illiquid commodities. 3.4 Robustness Test - Influence of Crude Oil Earlier in the study we pointed out that Crude oil market accounts for 20%-30% of the entire commodity market in terms of value. Therefore, one might argue that Crude oil may be driving our results on liquidity and hedging premiums in illiquid commodities. Moreover, we find evidence of Crude oil price spillovers on other commodities (Baffes 2007) and spillover effects are significant especially after the Global Financial Crisis (Wang, Wu & Yang 2014). In order to test whether our results are dependent on Crude oil, we exclude Crude oil from our sample 13 Daskalaki & Skiadopoulos (2016) also find that margins and futures returns are negatively correlated. This could happen as a risk management strategy implemented by market financiers to minimize the default rate of traders 20

21 and re-estimate key regressions. Results are reported in table 9. These results confirm our earlier findings that; 1) speculators (hedgers) pay an additional premium to hedgers (speculators) for liquidity (insurance) in illiquid commodities, 2) Speculators demand a higher premium to provide insurance in small size commodities markets and 3) Speculators pay a larger premium in bear market for liquidity in most illiquid commodities while the insurance premium paid by hedgers is not affected by neither bullish nor bearish market trends. Therefore, we conclude that our results are robust to the exclusion of Crude oil. 4 Conclusion In this paper we extend the work of Kang et al. (2017) to examine the behavior of liquidity and insurance premiums when it comes to illiquid commodities. Using the Amihud (2002) measure to capture illiquidity, we find that premiums capturing liquidity and insurance, explains commodity futures returns. However, Cochrane (2005), among others suggest that the Amihud measure is largely influenced by the size of the asset and therefore it makes more sense to consider the turnover version of the Amihud measure, that is the absolute returns per unit turnover, as the measure of liquidity (Florackis et al. 2011) and (Brennan et al. 2013). In support of this, we find a strong negative correlation between the market-size-weighted average Amihud measure and total market size, especially after This motivates use of turnover version of the Amihud measure, to capture illiquidity. Our analysis shows that there is an additional illiquidity premium attached to both liquidity and insurance, which means that speculators (hedgers) should pay to buy liquidity (insurance) in illiquid commodities, from hedgers (speculators). Once the Amihud measure is decomposed in to turnover-based liquidity and size, we observe that there is a cost to investing in small sized commodities. According to our findings, this cost is only associated with the insurance premium that hedgers pay to protect themselves against future price fluctuations. Price paid by speculators to secure short-term liquidity is not affected by the size of the individual commodities markets. Following the footsteps of Brennan et al. (2013), we examine the importance of the additional premium that speculators/hedgers pay for illiquid commodities in bull (up) and bear (down) markets. We observe that hedgers demand a significantly larger premium to provide short-term liquidity to trade illiquid commodities in down markets. However, speculators do not demand an additional insurance premium depending on the market condition, in order to 21