Predicting the Unpredictable : New Experimental Evidence on Forecasting Random Walks

We investigate how individuals use measures of apparent predictability from price charts to predict future market prices. Subjects in our experiment predict both random walk times series, as in the seminal work by Bloomfied and Hales (2002) (BH), and stock price time series. We successfully replicate the experimental findings in BH that subjects are less trend-chasing when there are more reversals in the first task. We find that subjects also overreact less to the trend when there is less momentum in the stock price in the second task, though the momentum factor that is significant is the autocorrelation instead of the number of reversals per se. Our subjects also appear to use other variables such as amplitude and volatility as measures of predictability. However, as random walk theory predicts, relying on apparent patterns in past data does not improve their prediction accuracy