Accurately predict the probability of the rise or fall of 2,600 stock prices, in bulk, three months in the future, to feed a portfolio investment/divestment process.
Standard and Poor's needed high accuracy rankable estimates of the rise and fall of a large volume of stock prices for a fund advisory portfolio management system. We provided the "Enterprise Modeling Server" (EMS III) neural network forecasting system as a solution. This solution builds, validates and rebuilds as needed 10's, 100's or 1,000's of genetically optimized neural network models. The EMS system had been proven effective previously for large volume customers such as DisneyWorld. Standard and Poor's provided proprietary inputs which were, in part, based on their Fair Value metrics. The 2,600 stocks were selected from the much larger universe of stocks because they had a long enough history (not new issues) and had not recently gone through a significant merger (notable behavior change). The data provided was in a relational database that the EMS extracted from using parameterized queries, the data was then modeled in a queued parallel processing walk-forward batch process, then the models were validated on hold-back data and failing models were requeued automatically to be rebuilt. This process was repeated until satisfactory models were obtained or the stock was put into an exception queue. Forecasting was then performed out over a 3 month horizon and the probabilities of the rise and fall created by the models were written to a relational database. The portfolio building process provided by Standard and Poor's extracted these forecasts, ranked them, "tiled" them (pentile or decile), and divested from stocks expected to fall (bottom pentile or decile) and invested in the ones expected to rise (top pentile or decile).
The system performed very well.
- Title: High Volume Securities Forecasting
- Client: Standard and Poor's
- Challenge: Customer wants accurate predictions of the rise and fall of securities for portfolio management.
- Skills: Data Access, High Volume Modeling, Batch Prediction, Performance Monitoring and Remodeling
Accurately predicted in bulk three months in advance the probability of the rise or fall of 2,600 stock prices to supply the estimates into deciled portfolio investment/divestment.