A test of the accuracy of the Lee/Ready trade classification algorithm

We analyze the accuracy of the Lee/Ready (1991) trade classification algorithm and the tick test. Our definition of true trade classification is based on whether the Makler (the equivalent of the specialist on the Frankfurt Stock Exchange) bought or sold shares. The Lee/Ready method classifies 72.8% of the transactions correctly. The simpler tick test performs almost equally well. We document that misclassification of trades may systematically bias the results of empirical microstructure research. Finally, we show that estimation of the bid–ask spread from transaction data results in a reasonably accurate estimate of the relative liquidity of our sample stocks. This is an important finding because quote data for the German stock market is not available on a regular basis.     

On the occurrence and consequences of inaccurate trade classification

The validity of many economic studies hinges on the ability to properly classify trades as buyer or seller-initiated. This study uses the TORQ data to investigate the performance of the Lee and Ready (1991, Journal of Finance 46, 733–746.) trade classification algorithm. I find that the algorithm correctly classifies 85% of the transactions in my sample, but systematically misclassifies transactions at the midpoint of the bid–ask spread, small transactions, and transactions in large or frequently traded stocks. I then provide evidence of the biases induced by inaccurate trade classification.     

Trade classification algorithms for electronic communications network trades

Ellis et al. [Ellis, K., Michaely, R., O’Hara, M., 2000. The accuracy of trade classification rules: Evidence from Nasdaq. Journal of Financial and Quantitative Analysis 35 (4), 529–551] find that trade classification rules have limited success in classifying trades which execute inside the quotes. We reconfirm this result and propose an alternative algorithm to improve the classification accuracy for trades inside the quotes. This alternative algorithm improves the overall success rate for classifying trades, especially for trades that occur inside the quotes. Additionally, we show that the Lee and Ready [Lee, C., Ready, M., 1991. Inferring trade direction from intraday data. Journal of Finance 46, 733–747] and Ellis et al. (2000) trade classification algorithms provide biased estimates of the actual effective spreads and price impacts, while our algorithm provides statistically unbiased estimates of actual effective spreads and price impacts.     

Short sales,long sales,and the Lee–Ready trade classification algorithm revisited

Asquith et al. (2010) conclude that short sales are often misclassified by the Lee–Ready algorithm. The algorithm identifies most short sales as buyer-initiated, whereas the authors posit that short sales should be overwhelmingly seller-initiated. Using order data to identify true trade initiator, we document that short sales are, in fact, predominantly buyer-initiated and that the Lee–Ready algorithm correctly classifies most of them. Misclassification rates for short and long sales are near zero at the daily level. At the trade level, misclassification rates are 31% using contemporaneous quotes and trades and decline to 21% when quotes are lagged one second.     

Signing trades and an evaluation of the Lee–Ready algorithm

We use techniques developed to analyze the Supply Curve in liquidity models in order to analyze the accuracy of the Lee and Ready algorithm, both for highly liquid and relatively liquid stocks. Through the use of order book data combined with tick data, we are actually (somewhat tediously) able to tell whether or not a given trade is buyer or seller initiated. For those trades where such knowledge is certain, the accuracy of the Lee and Ready algorithm is not as accurate as has been assumed previously. We can essentially prove that the Lee and Ready algorithm is always at least 55% accurate, and is around 61% accurate for highly liquid stocks (i.e., the top 50 of the S&P 100).     

Trade signing in fast markets

This study assesses the accuracy of trade signing algorithms in fast trading environments using NASDAQ and NYSE trade and quote data. Using data that contain true trade signs, we show that the Lee and Ready algorithm outperforms the tick rule and classifies trades at least as well as in earlier studies from slower trading environments, even in subsamples where the market is particularly fast. We conclude that trade signing remains viable in fast markets, and that the use of quote data continues to increase trade classification accuracy.     

The accuracy of the tick test: Evidence from the Australian stock exchange

In the absence of information regarding whether a trade is buyer or seller initiated, many researchers have employed the ‘tick’ rule as a proxy. These researchers have been supported in their endeavours by the work of Lee and Ready (1991) which suggests that the tick rule is 90% accurate. Unfortunately, the difficulty of securing data on this issue has made Lee and Ready's paper somewhat unique in that there have been few attempts to confirm their result in US markets and no attempts in other markets. The purpose of this work is to test the robustness of their result in the Australian securities market. Using cleaner intra-day data we mimic the Lee and Ready study to cast some doubt upon the robustness of their findings in different markets. Our results suggest an overall accuracy of approximately 74% as opposed to Lee and Ready's 90%. However, accuracy in excess of 90% is documented when zero ticks are excluded. Further analysis provides evidence that a volatile or trending market will decrease the accuracy of the tick rule. It is also demonstrated that the tick rule is less likely to accurately classify seller initiated trades and small buyer initiated trades.     

Dealer behavior and trading systems in foreign exchange markets

We study dealer behavior in the foreign exchange spot market using detailed observations on all the transactions of four interbank dealers. There is strong support for an information effect in incoming trades. The direction of trade is most important, but we also find that the information effect increases with trade size in direct bilateral trades. All four dealers control their inventory intensively. Inventory control is not, however, manifested through a dealer's own prices in contrast to findings by Lyons (J. Financial Econ. 39(1995) 321). Furthermore, we document differences in trading styles, especially how they actually control their inventories.     

NYSE order flow,spreads, and information

This paper uses unique NYSE audit trail data to evaluate spreads and information content for different order types. Actual spreads are positive for liquidity-demanding orders and negative for liquidity-supplying orders after controlling for order direction. However, because a large fraction of liquidity-demanding orders get price improvement, the actual spread for liquidity-demanding orders is up to 50 percent less than the Lee and Ready (1991) algorithm would suggest. Regression results show that the order composition of trades affects traditional measures of spreads and information. They also show that NYSE non-displayed liquidity reduces trading costs facing market orders, and that liquidity-demanding floor broker orders are the most informative order type.     

1-share orders and trades

1-share trades are the most common odd lot trade size, accounting for 9.62% of all odd lot transactions and 3.65% of all trades on NASDAQ in 2012. While 50.41% of 1-share trades result from broken orders, 34.89% of 1-share trades are intentional. We provide substantial evidence that traders use 1-share trades to “ping” for hidden liquidity. In particular, our results indicate that 1-share trades are disproportionately aggressive and also execute against hidden liquidity more than any other odd lot trade size. We also find a relative increase in trading immediately following a 1-share trade. Our results are in line with Clark-Joseph (2014), who suggests that traders may use small, unprofitable trades to detect information from other traders. Specifically, 1-share trades represent the minimum cash outlay necessary to trade, while simultaneously producing the smallest possible effects on a market maker's inventory, and in turn, a security's price.     

Transactions, volume, and volatility

We show that the positive volatility-volume relation documentedby numerous researchers actually reflects the positive relationbetween volatility and the number of transactions. Thus, itis the occurrence of transactions per se, and not their size,that generates volatility; trade size has no information beyondthat contained in the frequency of transactions. Our resultssuggest that theoretical research needs to entertain scenariosin which (i) both the frequency and size of trades are endogenouslydetermined, yet (ii) the size of trades has no information contentbeyond that contained in the number of transactions.     

The impact of transaction duration,volume and direction on price dynamics and volatility

We explore the role of trade volume, trade direction, and the duration between trades in explaining price dynamics and volatility using an Asymmetric Autoregressive Conditional Duration model applied to intraday transactions data. Our results suggest that volume, direction and duration are important determinants of price dynamics, while duration is also an important determinant of volatility. However, the impact of volume and direction on volatility is marginal after controlling for duration, and the impact of volume on volatility appears to be confined to periods of infrequent trading.     

Evaluation of the biases in execution cost estimation using trade and quote data

We use order data to assess the accuracy of execution cost estimation with trade and quote data. For our sample, estimates of the effective spread overstate actual execution costs by up to 17%. The biases result from errors in the inference of the trade direction and errors in the assignment of the benchmark quote. We find the accuracy of two popular trade direction algorithms improve marginally when trades are not lagged 5 seconds. Evaluation of the biases in execution cost measurement reveal the Ellis et al. (Journal of Financial and Quantitative Analysis (2000) 529) trade direction algorithm, combined with assigning benchmark quotes contemporaneous with trades, provides the least amount of bias. In general, biases are lower for relative effective spread estimates than effective spread estimates.     

Short sales and trade classification algorithms

This paper demonstrates that short sales are often misclassified as buyer-initiated by the Lee–Ready and other commonly used trade classification algorithms. This result is due in part to regulations which require that short sales be executed on an uptick or zero-uptick. In addition, while the literature considers “immediacy premiums” in determining trade direction, it ignores the often larger borrowing premiums that short sellers must pay. Since short sales constitute approximately 30% of all trade volume on U.S. exchanges, these results are important to the empirical market microstructure literature, as well as to measures that rely upon trade classification, such as the probability of informed trading (PIN) metric.     

Anonymity, Adverse Selection, and the Sorting of Interdealer Trades

This article uses unique data from the London Stock Exchangeto examine how trader anonymity and market liquidity affectdealers' decisions about where to place interdealer trades.During our sample period, dealers could trade with each otherin the direct, nonanonymous public market or use one of fouranonymous brokered trading systems. Surprisingly, we find thatadverse selection is less prevalent in the anonymous brokeredmarkets. We show that this pattern can be explained by the waydealers "price" the adverse selection risk inherent in tradingwith other dealers. We also relate our findings to recent changesin dealer markets.     

PIP Transactions,Price Improvement,Informed Trades and Order Execution Quality

This study focuses on innovations in order execution processes within the context of the Boston Option Exchange (BOX). More specifically, it examines the impact of the Price Improvement Process (PIP) on options quoted, effective and realised proportional spreads. We consider the PIP as a mechanism that allows the market maker to ‘internalise’ the transaction. We show that PIP transactions are associated with wider bid/ask proportional quoted spreads than non‐PIP transactions, in spite of the temporary narrowing of the effective proportional spread during PIP. We identify informed traders by focusing on the direction of trade. Using an original data set, we show that PIP transactions follow signals in the form of buy/sell orders by informed traders. We also show that PIP is a mechanism that allows the market maker to internalise a position in the same direction as that of the informed trader. We conclude that PIP does not improve the efficiency of the market but simply allows the market maker to benefit at the expense of uninformed traders.     

Transaction costs in an illiquid order‐driven market

Transaction costs in many international equity markets are much larger than those in the USA. This raises questions such as what trade size these reported trading costs relate to and whether investors can reduce trading costs by timing their trades. We show, using data from the order‐driven New Zealand market, that transaction costs are frequently lower for larger trades, particularly in small stocks, and investors are able to reduce costs by timing their transactions. While investors who require immediate execution incur transaction costs that are much higher than reported average costs, patient investors can trade at much better rates.     

Optimal trading strategy and supply/demand dynamics

In this paper, we study how the intertemporal supply/demand of a security affects trading strategy. We develop a general framework for a limit order book market to capture the dynamics of supply/demand. We show that the optimal strategy to execute an order does not depend on the static properties of supply/demand such as bid–ask spread and market depth, it depends on their dynamic properties such as resilience: the speed at which supply/demand recovers to its steady state after a trade. In general, the optimal strategy is quite complex, mixing large and small trades, and can substantially lower execution cost. Large trades remove the existing liquidity to attract new liquidity, while small trades allow the trader to further absorb any incoming liquidity flow.     

How and when is dual trading irrelevant?

Within a general model of speculative trade, we derive the aggregate consequences of dual traders who process retail liquidity trades and trade on their own account. We prove that dual trading reduces total expected speculator profits unless speculators process all liquidity trade and trade with the same intensity on liquidity trade. In contrast, dual trading does not affect the information content of prices. We show how results generalize when we endogenize (a) speculator information via costly information acquisition about fundamentals or costly processing of liquidity trade, and (b) liquidity trader motives and welfare via endowment shocks.     

Trading and enforcing patent rights

We study how the market for innovation affects enforcement of patent rights. We show that patent transactions arising from comparative advantages in commercialization increase litigation, but trades driven by advantages in patent enforcement reduce it. Using data on trade and litigation of individually owned patents in the United States, we exploit variation in capital gains tax rates across states as an instrument to identify the causal effect of trade on litigation. We find that taxes strongly affect patent transactions, and that trade reduces litigation on average, but the impact is heterogeneous. Patents with larger potential gains from trade are more likely to change ownership, and the impact depends critically on transaction characteristics.