Project: Microsoft's closing stock price prediction by ARIMA, Decision Tree and GARCH models in R Studio

Data Source: https://www.kaggle.com/datasets/vijayvvenkitesh/microsoft-stock-time-series-analysis

Approach

• Three prediction models: ARIMA, Decision Tree and GARCH are used to fit and predict on Microsoft's closing stock price data from 04-01-2015 to 3/31/2021.
• The trend of stock price is visualized to identify patterns and then Stationarity check is performed with ADF test.
• The stock time series is found to be non-stationary and thus using log transformation it is made stationary.
• Jarque Bera test is used to test normality and it is found that the time series is not approximately normal.
• Train-test splitting: Data before 1st Feb 2021 is train set and rest data for approx 2 months is used as test set.
• Autocorrelation and Partial autocorrelation testing performed with ACF and PACF plots.
• ACF plot shows evidence of autocorrelation as spikes for different lags go outside of blue region i.e. stock price in the present day depends on the past days.
• PACF plot shows no evidence of partial autocorrelation as only one significant spike is observed at 0 lag and for the rest lags the spikes are within the blue region.
• ARIMA model parameters tuned automatically with auto.arima() function. Optimal hyperparameters: p = 1, d= 1 and q=0.
• Decision tree model parameters tuned manually. Optimal hyperparameters: max depth = 10, min samples required for split = 4, min bucket size = 2 and rest kept as default.
• GARCH model tuned manually. Optimal hyperparameters: variance model= 'eGARCH' with p = q = 1, mean model = 'ARMA' with p = 2, q = 5 and rest kept as default.
• All models are tuned to minimize MSE and then tuned models are evaluated in test set in terms of RMSE, MAE and MAPE.

Performance comparison

FIndings and Recommendations

• Microsoft stock is increasing rapidly in the period 2020-21 and a similar or even better rise is expected in the near future from the given period.
• The lowest scores of metrics in the test set are produced by the decision tree as supported by line chart of actual and predicted close prices.
• Investors at present time are recommended to invest in Microsoft stock to gain significant profit.
• To precisely monitor Microsoft stock the decision tree model can be used by investors or shareholders to gain some rough idea about the trend in the future.
• Forecasting in distant future from the last date in sample data is not recommended as even the best model, decision tree would predict significantly inaccurate prices.