International Commodity Prices and the Australian Stock Market


  •  The authors thank Renee Fry, Mardi Dungey and other participants at the workshop on financial econometrics, University of Tasmania (2009).

George Milunovich, Department of Economics, Macquarie University, North Ryde, NSW 2109, Australia. Email:


We propose a method for estimating the earliest time during the trading day when overnight information is reflected in domestic share prices, and use it to measure the impact of international commodities on four Australian Securities Exchange (ASX) indices. While evidence is found that the ASX opening price does not fully reflect overnight news, this information is absorbed within 15 min of the opening time. Using appropriately constructed returns, we find international commodities to have a statistically significant and economically meaningful effect on the ASX. Nevertheless, the S&P 500 index appears to be a more important contributor of relevant overnight information.

I Introduction

The significance of the mining and agricultural industries for overall economic activity in Australia is well known and long standing. These industries also play an important role on the Australian Securities Exchange (ASX), representing between 25 and 30 per cent of the ASX capitalisation. Although several domestically produced commodities are traded on the Sydney Futures Exchange, major commodity prices are set in large international markets. Surprisingly, however, there are few studies that investigate the impact of globally determined commodity prices on the ASX listed firms.1

In this article, we investigate the extent of overnight dependence of the ASX returns on three international commodity indices: energy, metals and agriculture, which we construct from 29 commodities traded in the United States, the United Kingdom and Canada. To control for the effect of overnight economy-wide news, we also include the S&P 500 market-wide index as an explanatory variable. We propose that the degree of dependence of the Australian share market on overnight commodity news is best measured using an ASX price that is nearest in time to the opening price, but which reflects all overnight information available at the ASX opening time. If the presence of market frictions implies that the opening ASX price does not instantly and fully reflect the backlog of overnight information, we may underestimate its degree of dependence using the official opening price. Recent evidence of stickiness in the opening prices of major share markets has been reported in Baur and Jung (2006) and Milunovich and Thorp (2007). To eliminate the possibility of this downward bias, we identify the earliest point in time, to the nearest 15 min, when the ASX share price fully reflects information that has accumulated overnight from outside sources. The ASX price recorded at this point in time is called the ‘full-absorption’ price. Once a ‘full-absorption’ time is identified, we calculate overnight domestic returns, defined as log differences between the daily ‘full-absorption’ price and the previous day’s closing price. We then regress these overnight ASX returns on the previous calendar day daytime (close–open) commodity returns. As the ASX closing price used in this calculation is known at the point in time when the North American and UK commodity markets start trading, the R2 from these equations can be used as an estimate of dependency of the ASX ‘full-absorption’ price on the overnight commodity news.

Conceptually our approach resembles the early studies of the significance of overnight foreign information in international stock markets, such as Becker et al. (1990, 1992) and Hamao et al. (1990). These papers decompose daily (close–close) returns into overnight (close–open) and daytime (open–close) returns, and demonstrate that the impact of overnight foreign information persists into the trading day beyond the initial opening time of the domestic market. The significance of calculating open–close and close–open returns was also highlighted by Martens and Poon (2001) and Burns et al. (1998). These authors show that studies of market linkages can be significantly biased by non-synchronous trading problems and overlapping measurement of returns.

Our study extends this literature by developing a more robust high frequency econometric method, and incorporates information spillovers across different asset classes.

This article proceeds as follows: Section II explains the econometric methodology employed, while in Section III we discuss the dataset used. Section IV contains empirical results, and Section V concludes.

II Econometric Method and Data

Let St,it = 1,…,Ti = 0,…,24 represent the value of an ASX stock index i time periods after the opening time on day t, and measured in 15-min intervals. Also let Xt, t = 1,…,T be a 4 × 1 vector of overnight returns on three international commodity price indices (agriculture, metals and energy), as well as the S&P 500 index. As the US and the UK commodity markets open after the close of the ASX, and close before the next trading day begins in Australia, Xt−1 is known at the start of trading day t on the ASX. The first step in our empirical approach is to estimate the ‘full-absorption’ time i* when all overnight information is reflected in the ASX prices. To do this, we form a system of seemingly unrelated (SUR) regression equations as follows:


where αi is a scalar constant, inline imageis a inline image coefficient vector and ηit is a zero-mean error term. This set of 24 equations is augmented with one more regression:


which models the variation in the official opening index value.

The ASX operates a ‘pre-open’ period from 7 am to 10 am each trading day during which orders may be placed, but no trade takes place. Stocks with codes from A to B start trading at around 10:00:00 am; stocks with codes from C to F starting to trade at around 10:02:15 am, and so on. The last group of stocks to commence trading is the one with codes from S to Z, which start trading at around 10:09:00 am.2 The ASX calculates an opening price from the set of overlapping pre-open buy and sell bids, which minimises the absolute value of the excess demand.3 If this process works efficiently, then the opening price will reflect all the information about overnight commodity returns, and the subsequent information shocks will be uncorrelated with overnight information. In that case βi = 0 for i = 1,…, 24 in Equation (1). If the process is inefficient, however, βi will be non-zero for some low values of i, and will then become zero for higher values of i as the Australian share prices incorporate all overnight news. We define i* to be the smallest value of i for which βi = 0 when i > i*. In this case i* represents a point in time, following the official ASX opening time, after which the overnight news loses relevance for the subsequent ASX share returns.

A naïve approach to determining the impact of overnight commodity news on an ASX stock price index would be to regress the ASX close–open return on overnight commodity returns. That is, to estimate Equation (2). However, this approach assumes that the opening price of the ASX index fully reflects the impact of overnight news from commodity markets, and is therefore potentially misleading. A more general approach, which we pursue in this article, is first to estimate i*, and then to estimate




(i) Estimation of the Full-Absorption Time i*

Clearly the estimation of i* is critical. If too low a value is used for i*, then the ordinary least square (OLS) estimator of the impact of overnight commodity returns is likely to be biased downward. In contrast, if too high a value is used, then the OLS estimator of the impact of overnight commodity returns will generally be inefficient. This is most easily understood in the case where the intra-daily returns are assumed serially uncorrelated. In this case, whenever j > i*


where inline image is the OLS estimator of inline image for i = 0,…,24.

Ideally, a consistent estimator of i* should be used as this provides a large-sample rationale for the two-step procedure of estimating i* and then consistently estimating the parameters in Equation (3). If i* is inconsistently estimated, then the probability that the model used to estimate the full impact of overnight news is mis-specified will be non-negligible, even in the theoretical case of an infinitely large sample. To this end, we propose that i* be estimated using standard model selection techniques. For a given value of t, Equations (1) and (2) form a SUR model with 25 equations (6 h of trading divided into 15-min intervals, plus the previous-day-close to current-day-open time interval). A given value of i* implies a set of restrictions on the SUR model inline image. Furthermore, the sequence of restrictions corresponding to i* = 0,…,24 generates a sequence of nested models. Consequently, the problem of choosing i* corresponds to the problem of choosing the correct model from a set of 25 nested models. A standard approach to model selection problems of this type is to choose the model that minimises the Schwarz Criterion, SC(i). The key feature of this approach is that inline image is a consistent estimator of i* in the sense that inline image. Other approaches to the identification of i* are possible. Indeed, any likelihood-based selection criterion with a penalty function f(T) for which inline image and inline image provides a consistent estimator of i*. A popular alternative to the SC, which has this property, is the Hannan–Quinn criterion (HQC), and this might be used instead of the SC. It should be noted, however, that not every model selection criterion constitutes a consistent estimation procedure. In particular, the popular Akaike information criterion (AIC) will choose a time later than i* with non-zero probability as inline image and is therefore inconsistent. Similarly, any procedure which chooses i* based on the results of a sequence of hypothesis tests has a non-zero probability of choosing an incorrect value for i*, even as inline image (i.e. the probability of making a type I error) and may also be subject to a size distortion. Inconsistent order selection criteria often have other properties which may be attractive in certain applications. However, in our context, where we treat i* as a parameter to be estimated, consistency is the most important criterion, and so it dictates our choice of estimation procedure for i*. Nevertheless, we also conduct a sensitivity analysis of choosing i* based on three alternative selection procedures: HQC, AIC and a sequence of nested hypothesis tests.

Once the estimate of i* is computed, we estimate Equation (3) by OLS. The efficiency of OLS in this case is explained by Greene (2003).

The estimator of inline image provides a measure of the full impact of overnight commodity returns on the stock index. As the R2 from Equation (3) quantifies the proportion of the ‘full-absorption’ price variation that is explained by overnight commodity news, it provides a useful measure of the linear dependence of the ASX share price indices on the vector of international commodity returns.

III Dataset

We use 15-min data on four ASX indices: materials (XMJ), industrials (XNJ), energy (XEJ) and the market-wide S&P/ASX 200 index (XJO), covering the period from 21 July 2003 to 31 December 2009 (1633 trading days). Given that the trading day for the ASX starts at 10:00 am and finishes at 4:00 pm in Sydney, the 15-min sampling frequency yields 25 observations each trading day, including the opening price. The share price indices used here are market capitalisation weighted and include ASX listed companies that qualify for inclusion in the S&P ASX 200 share price index. In principle, sampling at a frequency higher than 15 min might be preferable, as it could provide a more detailed picture of the impact of overnight news on the domestic market. However, as we are using indexes, estimation bias caused by the non-synchronicity of constituent securities trading4 is an issue at high frequencies; 15-min sampling is a frequency at which the data are available that we regard as a good compromise between these two competing concerns.

The data on commodity prices consists of the opening and closing prices of spot and nearest-to-expiry commodity futures contracts traded in the international commodity markets. In total we collect prices from 10 commodity exchanges including the Chicago Mercantile Exchange, the New York Mercantile Exchange, the New York Board of Trade, the London International Financial Futures Exchange and the London Metals Exchange. A full account of the commodity exchanges used in this article is provided in Table A1 of the Appendix, together with a list of the commodities that we consider, and brief descriptions of the contract. Using the commodity contracts presented in Table A1, we construct three equally weighted commodity returns indices that correspond with agricultural, metals and energy commodities. An alternative approach would be to construct value- or volume-weighted commodity indexes. Unfortunately, the trade classifications used by the Australian Bureau of Statistics for commodities do not map well into the set of the contracts for which we have price data, and so the construction of satisfactory value- or volume-weighted commodity indexes was not possible. Furthermore, equally weighted indexes have been used routinely in the literature; see, for example, Bodie and Rosansky (1980), Fama and French (1987) and Gorton and Rouwenhorst (2006).

Daytime returns are also calculated for the S&P 500 market-wide index and included in each regression equation to control for the impact of macroeconomic shocks. We choose the S&P 500 index for this purpose for two reasons. First, given the design of our study, we need an index that does not overlap in trading hours with the ASX. This consideration rules out any indices that include stocks traded on any of the major Asian stock exchanges. Second, the United States is the last major market to close before the opening of the ASX. Consequently, the S&P 500 is a more recent and comprehensive measure of price-relevant overnight news, than indices from markets that lie further to the east and therefore close earlier, such as the United Kingdom and Europe. The ASX index data and exchange rates are provided by SIRCA, and the international commodity prices are sourced from Bloomberg. All returns are denominated in Australian dollars; Table 1 presents the usual summary statistics.

Table 1. 
Summary Statistics for the Period July 2003 to December 2009
 MeanMedianStandard deviationSkewnessExcess kurtosisJacque-Bera
  1. Notes: Summary statistics are presented for overnight returns, ln(open[t]/close[t − 1]), for the Australian indices and daytime returns, ln(open[t]/close[t]), for the international portfolios. The null hypothesis of the Jacque-Bera test is that the skewness and excess kurtosis are jointly zero. Under the null, the statistic is inline image-distributed.

ASX XEJ (energy)0.0130.0120.117−0.00626.51547808
ASX XJO (S&P/ASX 200)0.0170.0180.308−0.2347.3973736
ASX XMJ (materials)0.0510.0470.939−0.2376.7993158
ASX XNJ (industrials)0.0080.0100.269−0.2968.6495110
International agriculture−0.043−0.0350.834−0.3791.805261
International metals0.0790.1021.002−0.2263.268740
International energy−0.054−0.1042.1200.0390.70835
S&P 5000.0060.0341.047−0.3307.4773841

As shown by Table 1, all four ASX indices recorded a positive average overnight return over the sample period. The XMJ (materials) index exhibited the highest average growth rate, but also the highest level of volatility. The four indices exhibit negative skewness and excess kurtosis. Jacque-Bera tests rejected the hypotheses of normality with P-values that are zero to four decimal places. Of the four international indices, the metals index has shown the highest rate of growth, while the international energy index underperformed with a negative average return over the sample period. The negative growth rates present in the table are because of negative returns on the USD and GBP recorded over the sample period. With the exception of the energy variable, the foreign indices exhibit negative skewness and excess kurtosis, which the Jacque-Bera test shows to be highly statistically significant.

IV Empirical Results

The SC minimisation procedure explained in Section II produced inline image, for each of the four ASX indices. That is, we estimate that the earliest time at which all overnight commodity information is reflected in the ASX index is 10:15 am. To gauge the robustness of the results, we also estimated i* using the HQC, the AIC and a sequential hypothesis testing procedure.5 The HQC and the sequential hypothesis testing procedure both yielded the same results as the SC. AIC chose 10:45 am for XEJ, 11 am for XMJ and 4:00 pm for XJO and XNJ. Given these results we assume that 10:15 am is the correct ‘full-absorption’ time, and proceed to measure the amount of dependence of the ASX on the international commodity news.

Using these estimates of the ‘full-absorption’ time i* we then estimate Equation (3) for each ASX index and present the results in Table 2.

Table 2. 
Impact of Overnight Commodity News on Close–‘Full-Absorption’ Returns
Dependent variablesinline imageInternational commoditiesinline imageinline imageinline image
MetalsAgricultureEnergyS&P 500
  1. Notes: inline image corresponds to a point in time (in 15-min intervals) following the official opening price when overnight news is reflected in Australian Securities Exchange (ASX) prices; thus, inline image corresponds to 15 min after the official opening time. P-values are reported in brackets. inline image is the coefficient of determination for the regression with S&P 500 excluded, while inline image is the coefficient of determination for the regression in which the dependent close–‘full-absorption’ return variable has been replaced by the close–‘official open’ return.

S&P/ASX200 (XJO)10.0389 (0.112)−0.0337 (0.213)0.0421 (0.000)0.6438 (0.000)0.520.020.49
ASX XEJ (energy)10.0792 (0.032)−0.0484 (0.221)0.2009 (0.000)0.5805 (0.000)0.390.150.19
ASX XMJ (materials)10.2636 (0.000)0.0116 (0.791)0.1103 (0.000)0.8480 (0.000)0.440.090.42
ASX XNJ (industrials)1−0.0247 (0.280)−0.0361 (0.175)0.0141 (0.169)0.6369 (0.000)0.490.010.33

As illustrated by Table 2, overnight commodity returns have a statistically significant effect on the ASX ‘full-absorption’ prices. However, not all commodities matter, and not all sectors of the ASX are equally affected. Furthermore, as the inline image shows, dropping the S&P 500 market-wide index from the regressions significantly reduces the explanatory power of the model.

The S&P/ASX 200 (XJO) index does not appear to be strongly related to the international commodities markets. The explanatory power of the model of 52 per cent is reduced to only 2 per cent when the S&P 500 is excluded from the regression. The only statistically significant commodity in the ASX 200 equation is the index comprising international energy commodities. Judging by the inline image value of 49 per cent, which measures the explanatory power of the model for the official opening price, this index absorbs international news relatively quickly, with only 3 per cent of the ‘full-absorption’ price remaining to be explained over the next 15 min of trading.

Not surprisingly, international energy prices are statistically significant in the ASX energy index (XEJ) equation. Two other significant explanatory variables are the international metals index and the S&P 500 index, which itself accounts for 24 of the total 39 per cent explanatory power of the model. The XEJ index appears to be less efficient than the ASX 200. Only 19 per cent of the index value is explained at the official opening time compared with 39 per cent at the ‘full-absorption’ time recorded 15 min past the opening time.

International metals and energy prices, as well as the S&P 500 index have a statistically significant impact on the ASX materials index (XMJ). When the S&P 500 is excluded from the regressions, the commodity prices account for around 9 per cent of the variation in the ASX materials index, compared with 44 per cent explained by the full set of explanatory variables. The XMJ index absorbs overnight news quickly, with about 42 per cent of the total 44 per cent of the ‘full-absorption’ price explained at the official opening time.

The only relevant source of overnight news for the ASX industrial (XNJ) index appears to be the S&P 500; none of the three international commodity indices seem to be statistically significant at any conventional level. This is confirmed by the inline image value of 0.01, which measures the explanatory power of the model that excludes the S&P 500 return from the set of explanatory variables. The XNJ index is relatively inefficient in absorbing the overnight news, with only 33 of the total 49 per cent of the model’s explanatory power achieved at the official opening price, with the remaining 16 per cent accomplished over the next 15 min.

As a final illustration of the information absorption processes uncovered here, in Figure 1 we plot the two statistically significant regression coefficients for the ASX 200, with i* varying across 15-min intervals from 10 am to 4 pm.

Figure 1.

Parameter Evolution Over the Course of a Trading Day

Notes: Plots are provided for some of those parameters that are statistically significantly different from zero at a 5 per cent significance level in Table 2. Plots for remaining equations are available upon request. The solid lines depict estimated betas while the dashed lines represent the 95 per cent confidence intervals.

The graphs illustrate how the overnight news absorption process takes place, with the regression coefficients converging rapidly (within 15 min of the opening time) to their true values at the ‘full-absorption’ time, and hover around the true values for the remainder of the day.

V Conclusions

We investigate the extent of dependence of Australian share market indices on overnight international commodity news. To account for the possibility of market inefficiencies we develop a method for determining the length of time, following the market’s official opening time, that the domestic market takes to fully reflect relevant information generated in international markets overnight, while the domestic market was closed.

We find some evidence of stickiness in the ASX opening price, in the sense that most of the reaction to overnight stock returns occurs after the market has opened. However, the market appears to be quick in processing this information and the ‘full absorption’ of information is accomplished within the first 15 min of trading. When using the appropriately constructed ‘full-absorption’ returns, we find that overnight commodity returns have a statistically and economically significant impact on the ASX listed firms.

The most important international commodities are metals and energy, and the most affected sectors are ASX energy and ASX materials. In the case of the ASX energy index, overnight news is responsible for approximately 39 per cent of the variation in the ‘full-absorption’ price, with only 15 per cent accounted for by the international commodity variables. International metals and energy prices have a statistically significant impact on the ASX materials index, explaining 9 per cent of its ‘full-absorption’ price. When the overall ASX 200 market index is considered, commodities news accounts for 2 per cent of the of the model’s total explanatory power of 52 per cent; in the case of the ASX industrials of the model’s total explanatory power of 49 per cent, 1 per cent is accounted for by the international commodity news.

While our results provide some support for the claim that the Australian market is commodity-based, they also put this claim in perspective: the regression coefficients on international commodity returns are considerably smaller than the coefficients corresponding to the S&P 500. This suggests that other factors play an important role in determining the response of the ASX to overnight information.

Future research in this area may apply a similar methodology to other market indices and other types of foreign news. An interesting extension would account for informational spillovers that arise from either positive or negative news shocks.



Table A1. 
List of Commodities
CommodityMarketCommodity groupContract details
  1. Notes: CBOT, Chicago Board of Trade; CME, Chicago Mercantile Exchange; COMEX, the New York Commodities Exchange; NYBOT, the New York Board of Trade; NYMEX, the New York Mercantile Exchange; KCBOT, the Kansas City Board of Trade; WCE, the Winnipeg Commodities Exchange; ICE, the Intercontinental Exchange; LIFFE, the London International Financial Futures Exchange; LME, the London Metal Exchange.

SoybeansCBOTAgriculturalContinuous futures (nearest to expiry)
OatsCBOTAgriculturalContinuous futures (nearest to expiry)
Rough riceCBOTAgriculturalContinuous futures (nearest to expiry)
CornCBOTAgriculturalContinuous futures (nearest to expiry)
WheatCBOTAgriculturalContinuous futures (nearest to expiry)
Feeder cattleCMEAgriculturalContinuous futures (nearest to expiry)
Lean hogsCMEAgriculturalContinuous futures (nearest to expiry)
Live cattleCMEAgriculturalContinuous futures (nearest to expiry)
LumberCMEAgriculturalContinuous futures (nearest to expiry)
High-grade copperCOMEXMetalContinuous futures (nearest to expiry)
GoldCOMEXMetalContinuous futures (nearest to expiry)
SilverCOMEXMetalContinuous futures (nearest to expiry)
Sugar no. 11NYBOTAgriculturalContinuous futures (nearest to expiry)
Crude oilNYMEXEnergyContinuous futures (nearest to expiry)
Natural gasNYMEXEnergyContinuous futures (nearest to expiry)
PalladiumNYMEXMetalContinuous futures (nearest to expiry)
PlatinumNYMEXMetalContinuous futures (nearest to expiry)
WheatKCBOTAgriculturalContinuous futures (nearest to expiry)
CanolaWCEAgriculturalContinuous futures (nearest to expiry)
Western barleyWCEAgriculturalContinuous futures (nearest to expiry)
Brent crudeICEEnergyContinuous futures (nearest to expiry)
SugarLIFFEAgriculturalContinuous futures (nearest to expiry)
WheatLIFFEAgriculturalContinuous futures (nearest to expiry)
AluminiumLMEMetalOfficial cash market price
CopperLMEMetalOfficial cash market price
LeadLMEMetalOfficial cash market price
NickelLMEMetalOfficial cash market price
TinLMEMetalOfficial cash market price
ZincLMEMetalOfficial cash market price