Empirical research on the microstructure of exchanges ordinarily uses intraday price and trade data. However, the format and quality of the data differ greatly by exchange and data source. For measuring quoted spreads, a record of simultaneous bid and ask quotes is needed. In dealership or hybrid markets (with a designated specialist or market-makers with quote obligations), such data are sometimes available. However, one needs to be careful because the quotes may only be indicative (as they are, for example, in the foreign exchange markets) or there may be better prices available from a competing order book.

For the other measures discussed, prices of actual transactions are needed, and typically one also needs to know whether the transaction was initiated by the buyer (‘buy’) or by the seller (‘sell’). For some markets, this classification into buys and sells is easy to establish, since trading is at best quotes or the best prices available in the limit order book (LOB). An important exception is the NYSE, where trading can be within the specialist’s quotes, because either the specialist improves on his quoted price, or trading is with the LOB. Moreover, the NYSE data sources (notably the TAQ database) have two separate files for trades and quotes and the timing is not exactly simultaneous.

This sometimes makes it difficult to classify a trade as buy or sell. Lee and Ready (1991) developed a classification method that has become the standard measure for the TAQ data. In recent years, many high-quality data sets have become available from electronic limit order book markets, where all trades are cleared against the limit order book. This permits unambiguous classification of trades as buys or sells.