S&P 500 Predictions with Artificial Intelligence & Machine Learning

The US Economy has been growing for last 10+ years, and has recovered fully from 2008 recession. Global stock markets have also grown many folds. The S&P 500 has increased from 800 points in 2008 to 3000 points in early July 2019. With a mix of positive and negative news - increasing federal debt, increasing interest rates, inverted yield curve, shrinking federal balance sheet, strong consumer spending, mixed corporate results, trade wars etc, it is not easy to conclude whether the markets will continue to trend up or change course in near future. Investment advisors, with their experience in how markets performed in past, are making predictions based on only limited set of predictors.

Artificial Intelligence & Machine learning can help make these predictions using multiple predictors. This article highlights the results of my first experiment leveraging machine learning to make these predictions using multiple macro economic predictors. Prior to this experiment, I had unsuccessfully tried multi-regression models to establish the relationship between the predictors and the market index.

In this experiment with machine learning, I leveraged the economic data from last 50+ years, with number of direct and indirect predictors, like GDP, Inflation (CPI, PPI), Interest rates, Federal Reserve assets, Consumer Debt and many more, to estimate the S&P 500 index. The experiment uses machine learning to train a multi-layer recurrent network with historical time series of these predictors as input, to predict S&P in near future.

The model has successfully predicted S&P 500 at 3000 points around Jun/July 2019, along with other drops in the market. The variances in the recent predictions from the index could be attributed to the trade disputes, where the trained model lacks global trade predictors.

The trained model also seems to point towards a drop in the index in near future. Time will tell if the model is accurate, and whether it is the start of the next market contraction. I am still experimenting with the model, and don't recommend using it for any investment decisions. With the right level of computing power, predictors, and training - Machine learning does have the potential to predict future market trends.

The ideas and opinions presented in this article are my personal views. I am looking forward to getting some feedback from the AI community, and any one else working with similar economic models.