With this project I am introducing the design of a machine learning algorithm that predicts stock market crashes based on past price information. I used public available market data from seven major stock market indices. This is a classification problem to forecast whether or not a crash will occur within the next 1, 3 or 6 months at any point in time. I trained and tested linear and logistic regression models, support vector machines, decision trees and recurrent neural networks with long short term memory (RNN LSTM). For a detailed description of the problem and findings please refer to the article here
exploration.ipynb: Data exploration and crash definition and identification for all seven data sets.
results.ipynb: A summary of the results for training, validation and test results of all tested models. For more detail refer to the jupyter notebooks in the models directory.
models: Folder containing jupyter notebooks of all tested models. The notebooks illustrate how the models were trained, tuned and tested and present the results on the test set
data: .csv files of the daily price information of all datasets used. This data is available on yahoo finance.
final_predictor: Pre-trained logistic regression model that shows current forecasting probabilities for a crash occurring within the next 1, 3 and 6 months. Clone this repository, download most recent historical price information of any stock market from yahoo finance (at least 3 years of data), specify the filename in inputs.json and run main.py to get prediction results.
Example of crash prediction within 3 months on the S&P 500 (data used as test set) for the time between 1958 and 1976:
