Learning the True Index Level: Index Return Autocorrelation in an REE Auction Market

This paper develops and tests implications of cross-security information aggregation on index return autocorrelation. In the model, prices are realised individually and simultaneously in REE auction markets, then realigned to take information revealed in other prices into account. This adjustment is symmetric across stocks, leading to index return autocorrelation of MA (1) type. Autocorrelation will be high if the index level prior is noisy, for example, at Monday open and after high volatility in overnight trading. Autocorrelation will also be higher in portfolios of highly correlated securitites. Overnight information revelation and high trading volume reduces the noisiness of the index level prior and, consequently, return autocorrelation. Index return autocorrelation will be low, or even negative, if there is high cross-security correlation in revealed information, due to, for example, index arbitrage trading or profit taking. All major predictions are supported by tests using data from the Paris Bourse. In contrast to earlier models of index return autocorrelation, the model can generate both positive and negative index return autocorrelation. This paper also documents instances of negative index return autocorrelation.