DAILY TRADING VOLUME RELATED TO GOLD PRICE
DAILY TRADING VOLUME RELATED TO GOLD PRICE
James F. Feller, Middle Tennessee State University
The relationship of trading volume to asset return is a topic of interest both to the academician and to the professional trader. This paper examines that relationship using COMEX gold futures data and focusing on trading days characterized by abnormally large volume. The events tested indicate that price moves begin prior to excessive volume days, but that gold prices reverse shortly after, thus offering the possibility of contrarian trading.
The relationship between trading volume and the prices of both financial assets and commodities is of significant relevance to both financial economists and technical traders. The former group addresses the academic issues of information content and marker efficiency, while traders are primarily interested in the predictive value of volume within the mundane framework of profitable trading. Most generally, Schwager (1984) considers trading volume an indicator of a particular market’s liquidity. Shaleen (1991), on the other hand, regards volume as a measure of urgency either to buy or to sell. This paper is concerned specifically with trading in gold on the COMEX division of the New York Mercantile Exchange. Thus, volume is here defined as the number of gold contracts traded on the COMEX during each trading session. It should be noted that reported volume figures represent one side of a transaction. Daily total volume is equal to the total of contracts bought and the total sold. By construction, the number of long positions must always equal the number of short positions. As Nichols (1987) notes, COMEX is the largest gold futures exchange in the world, and thus its volume figures are the most important for traders. Volume figures, however, are not perfect indicators of market activity. Occasionally erroneous numbers are published in the financial press. Colby and Meyers (1988) note that volume is often erratic and unreliable. Distortions may result from holiday and seasonal patterns, buy-and-sell programs, dynamic hedging, arbitrage, and other market-related activities. Having tested several volume-based technical indicators, they conclude that such indicators generally yield poor trading results. Contrarily, Saitta (1998), using volume to confirm crossover signals of a simple moving average trading system, more than tripled the system’s returns in the bond futures market. Nichols (1987) attributes increases in volume to the actions of bullish speculators who react to rising prices by initiating long positions; similarly, increasing volume occurs with declining prices as these same large speculators liquidate their long positions, first with profit-taking, and later when threatened with margin calls. Low volume trading sessions, thus reflect the trades of only day traders and commercial hedgers. This explanation is incomplete, however, for the actions and motives of short sellers are not considered. Gero (1985) implies that increases in volume accompanied by rapid and large increases or decreases in price are attributable in the first instance to short covering, and in the second instance to increased short sales into the falling market. Further, he notes that increased short sales often precede an upside reversal. Volume, hence, should be interpreted considering the motives and market power of both buyers and short sellers of contracts. Finally, Edward and Magee (1966), Beale (1985), and Shaleen (1991) all concur that a price trend, to persist, must be complemented by a relatively high and moderately rising level of volume. A decline in volume, whether prices are in a bull or bear trend, often is associated with a trend reversal.
A technical indicator may be leading, coincident, or lagging relative to a given trend reversal. The leading indicator is of most value to a trader, but opinions as to whether volume leads or lags gold price moves are divided. The conventional wisdom, most tersely expressed by Murphy (1986) is: Volume precedes price. Schildgen (1986) and Pring (1980) support the conventional position with references to the works of several technical authors. Tvelde (1990) reverses the conventional sequence, and offers a psychologically-based explanation: as prices rise, traders take profits too quickly, and regret drives them back into the market, thereby increasing both prices and volume. Volume, then, follows and supports a bullish trend, and is strongest after an upside breakout. Moreover, Tvelde proposes that the direction of any price trend which follows a "sideways" movement, or trading range, of prices should be opposite from the direction in the trading range where the higher volume occurred. Touhey (1980), with regard to common stocks, also supports the price-precedes-volume position. The temporal relationship, if any, therefore remains undetermined.
The above discussion is especially noteworthy in that no academic works are cited. Indeed, there are few published papers which even approach the price-volume question relative to gold, and generally the issue is treated peripherally within the complex of efficient markets research. A number of studies deal with the price-volume relationship in non-gold markets. For example, Kamath, Ramchander, and Chandry (1998) determined that large absolute changes in price are positively related to changes in volume, and that ultimately the magnitude of change in volume affects the extent of the price movement. This conclusion, while illuminating, is based on an analysis of stock index (OEX) data, and thus may or may not be applicable to the gold market. The Kamath et al. study is only one of many such price-volume studies which relates to a specific market. Karpoff (1987), surveying the then existing literature, found numerous empirical studies supporting a positive correlation between absolute price change and daily volume in both financial and commodity markets. Karpoff did not, however, address the temporal sequence issue; rather, he limited his discussion to competing explanations of the price-volume relationship, emphasizing the potential alternative patterns of information flow. In his summary he set down eight directions for future research, all within the efficient markets paradigm. Frank and Stengos (1989) rigorously addressed the question of gold market efficiency. While they determined that the return-generating process is multidimensional, they did not suggest volume as one of the dimensions. Tauchen and Pitts (1983) considered volume a secondary source of information resulting from the number and intensity of information bits entering the market. While all of the above studies relate either to the gold market or to the price-volume relationship, none examines both phenomena simultaneously. Two recent studies do somewhat fill this vacuum. Chang, Pinegar, and Schacter (1997), using the most active COMEX gold contract data, related gold price volatility, as measured by Parkinson’s extreme value volatility measure, to total trading volume. With volume as the independent variable, their regression yielded a strongly positive relationship at the .05 level. Chang et al. concluded that the higher volatility associated with higher volume is the result of large speculators entering the market and aggressively opening new positions. Further, they set down two possible causes for such entry. First, they suggested that speculators might be drawn into the market after having observed an increase in volatility. Alternatively, speculators having superior information would enter the market to trade on that information, thus increasing both volatility and volume. Chang et al. did not, however, favor either explanation. Dhillon, Lasser, and Watanabe (1997) subsequently compared volume and volatility on the COMEX and TOCOM (Tokyo) gold futures markets. They found both trading day volume and volatility much higher on the much larger COMEX market, as expected. However, they also found the close-to-open volatility to be much higher on the TOCOM. Dhillon et al. concluded that more information is released during COMEX trading hours, and this information is conveyed to the Tokyo market during non-trading hours. Thus, volume effects on volatility result from the use of private information by large speculators. This conclusion, then, supports the "superior information" explanation advanced by Chang, Pinegar, and Schacter (1997).
The literature thus far examined from both technical and academic sources indicates a probable relationship between volume and price movements over multiple markets. Further, there is weak evidence to indicate that changes in volume result from differential information flows, rather than from changes in price volatility. None of the previously cited works, however, addresses the case of trading sessions characterized by abnormally high volume. Such periods, according to the Commodity Trading Manual (1989), are significant in that they signal major highs and lows, and the impending price reversals. Tauchen and Pitts (1983) attribute such climax volumes to high-intensity events external to the market, while Sudstrom (1992) associates them with the poor market timing of individual speculators. Edward and Magee (1966), Shaleen (1991), and Hutson, Weis, and Schroeder (1991) attribute such "blowoff" volumes to the panic selling of speculators who urgently need to liquidate losing positions. Pistolese (1993) and Murphy (1986) note that such abnormally high volume occurs both at market tops and at market bottoms, or selling climaxes. Both Allen (1972) and Schildgen (1986) set down a general scenario specific to the gold futures market. Just prior to a market top, volume first declines, then increases dramatically as the top is reached. Allen attributes the sharp increase in volume to new information, favorable to the uptrend, which triggers a bout of short covering. This short covering drives price higher in a self-reinforcing process as margin calls resulting from the higher price force more buying. By the end of this blowoff phase, both price and volume have risen to unsustainable levels, and both collapse owing to a lack of buyers. The short sellers will have liquidated their positions at a loss, and the bullish traders likewise will have taken their profits. Both Allen and Schildgen note that a reverse process occurs during washouts, or market bottoms. Further, following a market top or bottom accompanied by abnormally high volume, volume tends to decline as the price trend reverses.
Data and Method
The data examined in this study include all COMEX daily spot gold prices and trading volume figures reported by the financial press for the March 19,1986 to December 10, 1998 period. This initial sample yields 3,205 daily raw price returns calculated on a close-to-close basis; thus, overnight, weekend, holiday, and day-of-the-week effects are ignored. Somewhat fewer than half of the 3,205 volume figures, those recorded prior to April 24, 1992, have been rounded to the nearest 100 contracts. Considering the method employed here and described below, any effects arising from these measurement errors may be assumed trivial. Of more concern is the distribution of prices, and the derived daily gold returns. Solt and Swanson (1981), Aggarwal and Soenen (1988), Frank and Stengos (1989), and Poitras (1990) have found gold prices, price changes, and returns to be non-normally distributed. Return distributions tend to be leptokurtotic and positively skewed. Furthermore, time series are characterized by heteroscedasticity, and the distributions are non-stationary with respect to both mean and variance. Poitras (1990) noted that non-stationarity evolves over time as price levels change, and further observed that subsample distributions are "more normal." Solt and Swanson (1981) transformed gold price changes to logarithms, and found the presence of non-stationarity in the transformed distribution to be "less valid." Neter, Wesserman, and Kutner (1985) recommend a log transformation as a remedy for heteroscedasticity. Finally, by its nature, a log transformation tends to eliminate the moderate degree of positive skewness exhibited by return distributions. Thus, to mitigate any problems emerging from the non-normal character of gold price distributions, the raw data are converted to daily logarithmic returns.
The trading volume distribution also is non-normal, ranging from 2,289 to 182,284 contracts. The overall sample mean is 34,519, and the standard deviation is 18,214. The frequency distribution of raw volume figures exhibits an extreme positive skew, with approximately 78 percent of the values within one standard deviation of the mean. Transforming the data into one-period log relatives yields a reasonably symmetrical but leptokurtotic distribution. Especially apparent is the comparatively large number of exceptionally high values generated around holidays, when very low absolute values of volume are immediately followed by relatively higher values. By inspection, a frequency distribution of the logarithms of absolute daily volume figures yields the closest approximation to normality. While the volume figures are not directly involved in statistical testing, converting them into approximately normally distributed variates is appropriate insofar as they are used to derive the return subsamples which are tested. Schwager (1984), Edward and Magee (1966), and Shaleen (1991) emphasize that the absolute level of volume is of no significance. It is, rather, a particularly high volume, compared to the prevailing average daily volume, which is consequential. The definition of abnormal volume accordingly must involve some temporal average. Pring (1980) observes that, owing to the volatility of volume, some technical analysts use weighted averages to smooth the data. Shaleen (1991) suggests using a 30-day trailing moving average, but also points out the difficulties arising from the non-normality of the raw data.
Given the above discussion, the series of logarithms of the absolute volume figures was used in the subsample selection process. A 42-day trailing moving average of these data was constructed, along with the associated standard deviations and Z-scores for each date. Forty-two days is the average number of trading days each of the six major contracts of the calendar year is traded as the "nearby" contract. With the exception of the October contract, these contracts are the most actively traded until shortly before expiration when the next major contract becomes the "nearby" contract. The total sample was then sorted on the basis of Z-score, and the 20 dates having the highest relative volume were preliminarily selected. These sampled data were then sorted into chronological order, and the number of days between each succeeding date were calculated. Five of the 20 dates occurred within 42 days of the preceding date, and to ensure the independence of the subsample data, were eliminated. The question of long-term memory, or temporal dependence, in gold return series has been researched by, among others, Frank and Stengos (1989), Akgiray, Booth, Hatem, and Mustafa (1991), and Cheung and Lai (1993), but remains unresolved. Fifteen abnormal-volume dates were thus selected for testing, and are
Table 1. Sample Selected by Z-score and Date Elimination
summarized in Table 1, along with the associated total volume numbers and Z-scores. The calendar days between selected dates range from 83 to 690, or less than three months to nearly two years. Also of interest, eight of the 15 dates occur within three days of a month’s end, and ten, or 67 percent, occur in only three months: May, July, and November. Classifying the selected dates with regard to gold price movement is difficult in the enumerative sense. In terms of major price trend, eight of the critical dates fall within a bear market, six within a bull market, and only one, March 30, 1993, at an exact major turning point. Two other abnormal-volume dates preceded major price peaks, but by more than seven weeks in each case. Finally, the abnormal volume generated on May 23, 1990, occurred just three weeks prior to a significant market bottom and reversal to a bull trend. As to price direction on the abnormal-volume date itself, price rose significantly in ten instances, fell in four, and on May 28, 1992, moved only from $338 to $337. Further interpretation by inspection appears unwise, as an undesirable degree of subjectivism would likely bias any conclusions.
The method applied here involves a simple linear regression of daily gold returns on time. The time variables center on zero, the day of abnormal trading volume, and proceed from -20 through -1, representing the twenty preceding trading days, and 1-20 for the twenty following days. The general approach is labeled an event study, and is described on pages 428-431 in Elton and Gruber (1995).
Results and Conclusions
Regression results are summarized in Table 2. Most noteworthy, none of the slope coefficients is statistically significant, and all are near zero in absolute magnitude. The highest t-statistic, 1.64, is well below the critical value at the .05 level, 2.02, for a two-tailed test with 39 degrees of freedom. Further, the mean of the coefficients yields a near-zero value with a t-statistic of 16. The lack of explanatory power is also evinced by the low R2 values, of which nine are less than one percent. Given the random character of daily gold returns and the relatively short testing interval, the above results are not surprising. While the regressions represent the first step in this analysis, it is the calculation and averaging of the regression residuals that is more pertinent. A plot of the cumulative average residuals appears in Figure 1. During the twenty days preceding the abnormally-high-volume date, the cumulative average residuals fall to a minimum on the seventh trading day before the high-volume date, and then rise to a maximum on the fourth trading day following that date. The decline from the maximum then continues for another twelve trading days. These results indicate that abnormal-volume days occur, on average, seven trading days after a corresponding turning point in price, thus contradicting Murphy (1986), Schildgen (1986), and Pring (1980), and supporting the price-precedes-volume position advanced by Touhey (1980) and Tvelde (1990). When a five-day centered moving average is used to smooth the residuals series, the upturn begins six days prior the abnormal-volume date, and the series peaks on the third day after the volume surge. Of most interest, however, is the twelve or thirteen days which follow the peak, for this period of more than two calendar weeks represents a trading opportunity. Inspection of the data indicates that eleven of the fifteen sampled dates, or 73 percent, were associated with large positive price moves which reversed after a sharp peak, typical of a blowoff generated by panic-driven short covering. Three of the events, contrarily,
Table 2. Summary of Regression Results
Figure 1. CUMULATIVE AVERAGE RESIDUALS: RAW FIGURES
represented selling climaxes probably caused by the liquidation of long positions, while one abnormal-volume event occurred during a period of continuing lateral price movement. A simple trading strategy, thus, would involve the taking of a position contrary to the direction of the price movement. This position would be initiated three trading days after the abnormal volume date, and would be held open for approximately twelve trading days. As pointed out by Murphy (1986), volume figures are reported with a one day lag, and generally are not published in the financial press until two trading days have passed. Since the results here indicate that the abnormal volume day precedes a top by, on average, three days, the reporting lag is essentially irrelevant. A position trader depending only on hardcopy Wall Street Journals could enter trades based on an abnormal volume strategy.
This study suggests several tentative conclusions. First, the COMEX gold market is not efficient, insofar as an event generated within the market itself offers a potential for abnormal, albeit short term, returns. Second, the evidence presented here indicates that positive price movements precede abnormally large volume days, rather than the contrary. Third, support is offered for the supposition that large price movements accompanied by excessively high volume represent overreactions that are followed by reversals, rather than continuances. Fourth, lags in the reporting of abnormal volume figures do not appear significant with regard to trading on the basis of such reports. Finally, volume alone seems to convey important information about probable price movements; other technical variables, such as open interest, need not be considered in a short term trading strategy based on abnormal volume.
On the other hand, this study must be considered exploratory rather than definitive. While the total sample of over 3,000 return and volume data is adequate, only fifteen abnormal volume events were examined, and the majority of these were associated with positive price moves. Moreover, given the non-normality of the return distributions, the use of more-sophisticated regression techniques (e.g. GARCH) could be appropriate. In a practical sense, the suggested trading strategy was not actually tested, and trading costs, including slippage and commissions, were not considered. Similarly, if the strategy were operational, would futures or options on futures yield the better return? Finally, given a longer term trading perspective, the effects, both isolated and interactive, of other technical indicators, such as the previously mentioned open interest, would warrant examination. The potential for further research appears limitless.
References
Aggarwal, Raj, & Soenen, Luc (1988). The nature and efficiency of the gold market. Journal of Portfolio Management, 14, 18-21.
Akgiray, Vedat, Booth, G. Geoffrey, Hatem, John J., & Mustafa, Chowdhury. (1991). Conditional dependence in precious metal prices. Financial Review, 26, 367-386.
Allen, R. C. (1972). How to build a fortune in commodities (rev. ed.). Bright waters, NY: Windsor.
Beale, Robert (1985). Trading in gold futures. New York: Nichols.
Chang, Eric C., Pinegar, J. Michael, & Schacter, Barry (1997). Interday variations in volume, variance, and participation of large speculators. Journal of Banking & Finance, 21, 797-810.
Cheung, Yin-Wong, & Lai, Kon S. (1993). Do gold market returns have a long memory? Financial Review, 28, 181-202.
Colby, Robert W. & Meyers, Thomas A. (1988). The encyclopedia of technical market indicators. Homewood, IL: Dow Jones-Irwin.
Commodity trading manual (1989). Chicago: Chicago Board of Trade.
Dhillon, Upinder, Lasser, Dennis J., & Watanabe, Taiji (1997). Volatility, information, and double versus walrasian auction pricing in US and Japanese future markets. Journal of Banking & Finance, 21, 1045-1061.
Edward, Robert D., & Magee, John. (1966). Technical analysis of stock trends (5th ed.). Boston: John Magee.
Elton, Edwin J., & Gruber, Martin J. (1995). Modern portfolio theory and investment analysis (5th ed.). New York: Wiley.
Frank, Murray & Stengos, Thanasis (1989). Measuring the strangeness of gold and silver rate of return. Review of Economic Studies, 56, 553-567.
Gero, Anthony George (1985). Precious metals: A guide to trading and investing. Secaucus, NJ: Lyle Stuart.
Hutson, Jack K., Weis, David H., & Schroeder, Craig F. (1991). Charting the stock market. Seattle, WA: Technical Analysis, Inc.
Kamath, Ravindra, Chatrath, Arjun, Ramchander, Sanjay, & Chandry, Mukesh (1998). Market direction and volume-price variability relationships in speculative markets. American Business Review, 16, 20-24.
Karpoff, Jonathan M. (1987). The relationship between price changes and trading volume: A survey. Journal of Financial and Quantitative Analysis, 22, 109-126.
Murphy, John J. (1986). Technical analysis of the futures markets. New York: NYIF.
Neter, John, Wasserman, William, & Kutner, Michael H. (1985). Applied linear statistical models (2nd ed.). Homewood, IL: Irwin.
Nichols, Jeffrey A. (1987). The complete book of gold investing. Homewood, IL: Dow Jones-Irwin.
Pistolese, Clifford (1993). Nerves of steel. Chicago: Probus.
Poitras, Geoffrey (1990). The distribution of gold futures spreads. Journal of Futures Markets, 10, 643-659.
Pring, Martin J. (1980). Technical analysis explained. New York: McGraw-Mill.
Saitta, Alex (1998). A price and volume based system. Technical Analysis of Stocks & Commodities, 15 (13), 69-72.
Sundstrom, Roy E. (1992). Demystifying Wall Street. Chicago: Dearborn.
Schildgen, James E. (1986). Analytical methods for successful speculation. Chicago: Capital Futures Associates.
Schwager, Jack D. (1984). A complete guide to the futures markets. Wiley: New York.
Shaleen, Kenneth H. (1991). Volume and open interest. Chicago: Probus.
Solt, Michael E., & Swanson, Paul J. (1981). On the efficiency of the markets for gold and silver. Journal of Business, 54, 453-478.
Tauchen, G. & Pitts, M. (1983). The price-variability-volume relationship on speculative markets. Econometrica, 51, 485-505.
Touhey, John C. (1980). Stock market forecasting for alert investors. New York: AMACOM.
Tvelde, Lars (1990). The psychology of finance. Stavanger: Norwegian University Press.