A DECISION SUPPORT SYSTEM FOR MAXIMIZING TSP RETURNS WITH LIMITED FUND TRANSFER OPPORTUNITIES

Abstract

In this project, a decision tool is developed to manage Thrift Savings Plan (TSP) that is only allowed to trade twice a month. S&P 500 index data is downloaded from Yahoo finance and used to optimize the trading triggers as percentage stock changes (%in and %out). The Latin Hypercube Sampling (LHS) technique is used to obtain 2000 %in/%out combinations in the range of 1%-20%. Simulation results show that optimal %in and %out vary with length and time of stock data used. Thus, optimal %in/%out may not be transferrable from time to time, which considerably weakens the predictability of the tool. Stock returns improve if traders predict stocks 3-5 days into the future using the Long Short-Term Memory (LSTM) network. Adding an additional decision layer based on stock history improves S&P 500 returns under the 2 limit per month constraint but has little effect when unlimited trading is allowed. For long-term financial management such as TSP, two trades per month is sufficient, which theoretically proves the fairness of the 2 limits per month fund transfer rule set by the Federal Government. The simulated percentage of time in stocks matches the targeted percentage of fund in stocks for various Lifecycle Funds designed for different retirement ages in TSP. To improve the decision tool’s predictability, additional information from business, economics, government policy, and social media needs to be included and analyzed simultaneously and spontaneously.