In this paper, analyst Anton Linnér examined the potential of using artificial neural networks to create a portfolio strategy on the SPY index. The purpose was to utilize both price data and macroeconomic data in a neural network to predict the price movement of the SPY index in the coming week. The strategy was to use the weekly prediction to inform the decision on how to allocate between stocks and bonds until the next prediction was made.
The main aspects of this project were to find data relevant to price movements and to construct an appropriate neural network to handle this data. The main limitations of these aspects were the limitations of the magnitude of data and the prediction power of the data. Data pre-processing was therefore an important step in this project, in addition to constructing an applicable neural network for the task.
Using the neural network-directed portfolio strategy, a close to 70% return was obtained when backtesting on the SPY index during the time period 2018/06 – 2021/01. The buy and hold strategy had an almost 60 % return during the same time period, making the neural network strategy the more profitable option. However, even though the neural network strategy showed promising results, no significant conclusion can be drawn due to the short backtesting period and the particularity of the time period, such as the influence of a pandemic on the stock market.
To read the full report, please see attached PDF below:
