Estimating the probability of informed trading—does trade misclassification matter?

Abstract

Standard-Nutzungsbedingungen:Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden.Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen.Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in December 2002Abstract: Easley / Kiefer / O'Hara / Paperman (1996) (EKOP) have proposed an empirical methodology that allows to estimate the probability of informed trading and that has subsequently been used to address a wide range of issues in market microstructure. The data needed for estimation is the number of buyer-and seller-initiated trades. This information often has to be inferred by applying trade classification algorithms like the one proposed by Lee / Ready (1991). These algorithms are known to be inaccurate. In this paper we perform extensive simulations to show that inaccurate trade classification leads to biased estimation of the probability of informed trading when applying the EKOP methodology. The estimate is biased downward and the magnitude of the bias is related to the trading intensity of the stock in question. Scrutinizing prior empirical studies using the EKOP methodology, we conclude that the bias may severely affect the results of empirical microstructure studies.JEL classification: C52, G10, G14Keywords: Informed trading, market microstructure, trade classification * We thank seminar participants at the University of St. Gallen for helpful comments.

Short sales and trade classification algorithms

“…However, since EMO is less commonly used than the others, to save space, we do not report those results in this paper. 2 Boehmer, Grammig, and Theissen (2006) use post-decimalization data, but their focus is the impact trade misclassification has on the probability of informed trading (PIN) estimation. This paper applies the three commonly used trade classification algorithms to a new dataset of short sale transactions for stocks on the NYSE and NASDAQ in 2005.…”

Tick test accuracy in foreign exchange ECN markets

“…As noted by Theissen (2001), the accuracy of signing algorithms is critical. Boehmer et al (2007) and others note biases created by inaccurate signed trades, while Tangaard (2003) states that corrections or adjustments are not made even though most studies recognize the existence of signing errors. Perlin et al (2011) provides a stochastic model of the tick test that could be used to correct for errors, and we noted that the forex inventory effect of Lyons (1995) was not found by Bjonnes and Rime (2005) who used the Odders-White latest time stamp to sign forex trades instead of the tick test used by Lyons.…”

“…Theissen (2001) also notes that since most data sets do not identify or provide information about the trader, much less the initiator, the accuracy of the trade classification algorithm is crucial to the validity of empirical research. Boehmer et al (2007) finds that LR inaccurate trade classification causes significant downward bias in estimating the PIN (probability of informed trades), yet Tay et al (2009) use LR to estimate PINs without correcting for estimation bias. Although most research papers recognize the existence of error rates for these trade classification algorithms, Tanggaard (2003) points out that no corrections or adjustments are attempted for the potential bias in the misclassification of the buy/sell initiator.…”

“…One implication of their work is that the implied bias of the tick test can be evaluated even though the researcher only has trade data. This could address the consequences of trade sign misclassification (Tanggard (2003) and Boehmer et al (2007)). They also used trade data on 15 of the most liquid Brazilian stocks in 2009 to find that the tick test is about 72% accurate, which closely matched their model predictions.…”

“…Clearly, the tick rule avoids these potential matching errors. This matching issue may also affect the 'true' classification of buyer/seller initiated trades with proprietary data (see Boehmer et al (2007) and Bessimbinder (2003)), including ECN order data using the 'latest order submission time' method of Odders-White. Holden and Jacobsen (2011) examine the difficulties in matching trades to quotes with modern equity tick data.…”