In this era of high inflation and low interest rates, the public often invests in financial products to increase their passive income and thus increase their savings. Current data show that the public still considers stocks as investment targets. However, because invest-ment is risky and each person’s investment personality (risk tolerance) is different, some prefer to take risks to obtain the maximum return, and some are afraid of risks and avoid them to obtain stable returns; nevertheless, the recommendations of current research give little consideration to the risk characteristics of the investment target and the personal char-acteristics of the investor. However, investment is a trade-off between risk and reward, and the risk acceptance associated with an investment depends on the investor’s ability to ac-cept risk. Therefore, in the context of recommending, a consideration of investors’ per-sonal characteristics brings recommendations more in line with users’ expectations.
This study proposes a method to manage investment portfolios based on investors’ risk personalities. It is mainly used to classify stocks according to beta indicators (select stocks according to investors’ personalities), and financial indicators and an autoencoder are used to score stocks; moreover, technical indicators, covariance matrices, deep rein-forcement learning-advantage actor critic (A2C), and proximal policy optimization (PPO) are used for asset allocation with the aim of developing investment portfolios with good performance that are optimally suitable for different types of investors.
In the training results, the cumulative return rate obtained by the A2C model is as high as 49.61% for conservative investments, 82.04% for stable investments, and 99.69% for active investments. The cumulative return obtained by the PPO model is as high as 39.92% for conservative investments, 89.89% for stable investments, and 85.61% for ac-tive investments.