Prediction of Odd-Lot Stock Transactions

Operations-research problems sometimes involve the prediction of time series. An interesting process against which to demonstrate the prediction of multiple time series is the buying and selling of odd lots of stocks on the New York Stock Exchange. The process also offers the opportunity of observing how customer buying is affected by price changes. Examination of data published by the Securities and Exchange Commission reveals that customer response tends to lag behind market changes by about half a year, with customer behavior reflected in dealers' inventories with an additional three months lag. The capabilities of several linear prediction functions applied to dealers' inventory are compared, and a simple two point predictor is shown to be nearly as efficient as a more complex form.