6680268_​NintendoDailyStockPrice

Data Collection - Load Nintendo (7974.T) daily stock price data, Jan 5 - Jun 30, 2026 from Investing.com (119 observations) - Remove thousands separator from Price and convert to number - Convert Date column from String to Date and Time

Data Cleaning & Preparation - Sort data ascending by date (oldest to newest) - Remove unnecessary columns, only keeping Date and Price - Check for missing values (none found - 119 consecutive trading days) - Apply log transformation

Exploratory Visualization - Compare the effect of log transformation with the raw price Line Plot

Outlier Detection - Calculate 21-day rolling mean and standard deviation - Compute UCL and LCL - Flag any values outside UCL/LCL as outliers

Holt's Double Exponential Smoothing - Fit Level + Trend model on Log_Price - Optimize alpha and beta using brute force search - Evaluate using RMSE via Numeric Scorer Optimal: alpha = 0.3, beta = 0.3 RMSE = 0.034

Holt's Triple Exponential Smoothing - Fit Level + Trend + Seasonality model on Log_Price - Seasonal period m = 5 (trading week) - Optimize alpha, beta and gamma using brute force search - Evaluate using RMSE via Numeric Scorer Optimal: alpha = 0.2, beta = 0.6, gamma = 0.1 RMSE = 0.032

ARIMA Modeling - Split data 80/20 (95 training / 24 test) - d fixed at 1 based on ADF test - p and q selected by grid search over {0, 1, 2} - Train ARIMA Learner on 80% training set - Predict on 20% test set using ARIMA Predictor - Join forecast with actual test values, evaluate using RMSE Optimal order: ARIMA(1, 1, 2) RMSE = 0.037

Model Comparison - Compare RMSE of all three models: Holt's Triple = 0.032 Holt's Double = 0.034 ARIMA(1,1,2) = 0.037 - Holt's Triple performed best with the lowest error

Conclusion: Holt's Triple Exponential Smoothing performed best with the lowest RMSE of 0.032, followed by Holt's Double (0.034) and ARIMA(1,1,2) (0.037). Holt's Triple outperformed the others because it captures the weekly seasonal pattern (every 5 days) in Nintendo's daily trading data, while its low alpha (0.2) smooths the level and prevents overreaction to the high day-to-day volatility during a sharply declining market. ARIMA performed worst, suggesting that Nintendo's stock price is difficult to predict using past values alone. The differenced log series behaves close to white noise, meaning historical patterns carry little predictive power. Therefore, Holt's Triple is recommended for forecasting Nintendo's daily stock price.