The Return Predication and Risk Assessment of Tesla

Abstract

Tesla has emerged as a prominent corporation in recent years, driven by its development of innovative electric vehicles, including the Model 3, Model X, and Model Y. The company's core mission is to promote zero-emission transportation and accelerate the transition from fossil fuels to sustainable energy sources such as electricity. As a result, Tesla's stock has attracted significant attention from investors and experienced substantial appreciation. However, the onset of the COVID-19 pandemic led to delays in vehicle deliveries and broader economic downturn in the United States, contributing to a decline in Tesla’s stock price and a corresponding erosion of investor confidence. By 2023, Tesla’s business operations began to recover, prompting our team to develop multiple forecasting models to predict future stock prices based on historical volatility and relevant financial factors. This study aims to identify the optimal modeling approach that captures the relationship between future stock prices and associated risks. To achieve this objective, we propose three distinct models: Model I: Multi-Factor Model; Model II: Time Series Model; and Model III: Random Forest Model. Model I adopts a multi-factor framework grounded in the Fama-French three-factor model. The selected factors include Small Minus Big (SMB), which captures the size effect by comparing returns of small-market-capitalization firms with those of large-cap firms; High Minus Low (HML), which reflects the value premium by contrasting firms with high and low book-to-market ratios; and Momentum (Mom), which represents the "winner minus loser" effect, wherein stocks with strong past performance continue to exhibit superior returns in the near term. Model II employs an Autoregressive Integrated Moving Average (ARIMA) specification, which integrates autoregressive (AR) and moving average (MA) components. ARIMA models are widely used for short- to medium-term stock price forecasting by leveraging historical price patterns. The selected model, ARIMA (1,0,1), was determined after removing seasonality and confirming stationarity in the data, thus eliminating the need for differencing. Model selection was further validated using the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), alongside visual inspection of the Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF) plots of the stationary series. Model III utilizes a Random Forest regression algorithm, an ensemble learning method based on bootstrap aggregating (bagging). This technique involves constructing multiple decision trees from resampled datasets and employing random node splitting to enhance model robustness. Predictions are aggregated by averaging or weighted averaging across individual trees, yielding the final regression output. For sensitivity analysis, we compute rolling means and standard deviations of historical returns to simulate current-day returns. This approach enables the estimation of percentage changes over a moving window, thereby providing insight into the sensitivity of current returns to fluctuations in underlying market conditions.