عملکرد نسبی برآوردگرهای اسپرد قیمت خرید و فروش: شواهد از بازار آتی

The issue of transaction costs is the mainstay of the equity market microstructure. Research in the microstructure of futures markets has lagged behind. A primary reason is that futures exchanges in the U.S. do not record bid–ask quotes, requiring these costs to be imputed from transaction price data. A reliable estimator of bid–ask spreads would significantly enhance microstructure research in futures markets. Unique intraday data from the Sydney Futures Exchange (SFE) that include both transaction prices and bid–ask spreads allow us to compare bid–ask spread estimation techniques proposed in the literature against the benchmark of actual spreads in a futures market, and thus identify the best-performing estimator. To maximize relevance, we impose all the constraints that apply in U.S. futures data to perform our estimations. We find that the four bid–ask spread estimators considered significantly underestimate the actual spreads. However, simple moments-based estimators perform better in predicting spreads.

The microstructure of security markets influences investor, exchange, and regulator decisions. While microstructure issues have been studied extensively in equity markets, similar studies in futures markets are constrained by the absence of recorded bid–ask spreads in U.S. futures exchanges. Studies in these markets typically use an estimator to impute spreads from the price data available. We contribute to this research by comparing the performance of four commonly used spread estimators, Bhattacharya (1983), Roll (1984), Smith and Whaley (1994), and Thompson and Waller (1988), to actual spreads found in a market with a structure similar to U.S. futures exchanges, the Sydney Futures Exchange (SFE).2 This comprehensive analysis of the most commonly used estimators as well as actual spreads in a futures exchange makes our study unique in the futures market literature. Before its recent switch to electronic trading, SFE largely followed trading mechanisms used by major U.S. exchanges. We only use data before the advent of electronic trading, thus giving us a sample very similar to one obtained from exchanges in the U.S. To maximize further relevance, we impose all the constraints that apply in U.S. futures data to perform our estimations. Actual market quotes give us a benchmark for measurement of performance of the estimators. Thus, our empirical results are applicable to research using data from U.S. futures exchanges. Continuing advances in technology and a regulatory focus on enhancing competition have roiled global exchanges. Both these developments have significantly changed the competitive landscape, and nowhere have these changes been more apparent than in futures markets. Competition among exchanges most commonly occurs on the basis of costs and liquidity; markets with lower transaction costs are more efficient. Exchanges face a variety of decisions regarding optimal market structures. The success of any alternative choice will depend on how well it achieves the objectives of reducing costs and enhancing liquidity. The common measure of market liquidity and costs incurred by investors is the bid–ask spread. Research on the microstructure of futures markets is yet to catch up with equity market microstructure. One of the primary reasons is that the absence of recorded bid and ask quotes in U.S. futures markets, meaning we must impute transaction costs from price data.3 A reliable estimator of costs in terms of bid–ask spreads would significantly enhance participants’ ability to choose between different market structures. Actual quote data from SFE enable us to compare popular estimators against actual spreads in order to identify the most reliable estimator. We use intraday transaction prices and quotes to analyze these estimators and identify the one that most accurately describes the observed spreads. Contribution of our research to the literature is that it provides an assessment of the performance of these estimators by comparing the estimated spreads to the actual quotes in a floor-based market that is very similar to the U.S. futures markets. Besides comparing the accuracy of spread estimators, we analyze the errors in estimating bid–ask spreads. We use the determinants of spread as discussed by Harris (1994) to study the behavior of the errors. If the errors are randomly distributed, then we should find no relation between the errors and determinants of spreads as prices, volume, and volatility. All the bid–ask spread estimators we consider underestimate the actual spreads, and the errors are highly statistically significant in all cases. The Thompson–Waller and Bhattacharya estimators outperform the other two. Our results show that bid–ask spread estimates based on transactions data from floor trading are systematically biased. Contribution of our research is that this finding has significant implications for studies that compare floor-based and electronic futures market structures. Section 2 reviews the related microstructure literature. Section 3 discusses the bid–ask spread estimation methods and other related work. In Section 4, we explain the structure of futures markets, in particular that of SFE, and describe our data set in detail. Section 5 presents our empirical results. Section 6 concludes.

Bid–ask spreads are so important in market microstructure research that we must have reliable estimators for spreads if we are to have meaningful research on futures markets. We contribute by identifying the estimator that performs the best in predicting actual spreads. Our access to unique intraday data from SFE, which includes both transaction prices and spreads, allows us to compare the performance of four bid–ask spread estimation techniques against the benchmark of the actual spreads. We use data on two popular contracts, the Share Price Index futures and the Bank Accepted Bills futures. For both SPI and BAB futures, the Bhattacharya, Roll, Smith–Whaley, and Thompson–Waller bid–ask spread estimators underestimate the benchmark spreads. Two performance measurement criteria, mean absolute error and the correlation coefficient, allow us to capture the significant facets of the estimators. For both SPI and BAB futures, the three moment-based estimators (Thompson–Waller, Bhattacharya, and Smith–Whaley, in that order) outperform the negative autocovariancebased estimator proposed by Roll. While the Thompson–Waller and Bhattacharya estimators outperform the Smith–Whaley and Roll estimators by the measures used, all the estimators show significant errors in estimation and systematically fail to capture all aspects of the spread. The Thompson–Waller and Bhattacharya estimators represent the behavior of actual spreads in the market better than the Smith–Whaley and Roll estimators. Even so, little separates the two, and errors in estimating spreads are statistically significant and show systematic biases. Our findings show that bid–ask spread estimates based on transactions data from floor trading are systematically biased. This bias is likely to produce unreliable conclusions if studies compare estimated spreads of the pre-switch floor trading with the actual spreads of the post-switch electronic trading. In light of the move toward electronic trading in futures markets, investors, exchanges, and regulators are likely to be interested in more accurate bid–ask spread estimation methods. A more accurate bid–ask estimator can facilitate the comparison of transaction costs in two competing market structures which are floor-based and electronic trading.