Downloads 20 years of USDJPY data, fits LOWESS and polynomial/linear regressions, and plots historical prices with 10-year forecast and summary stats.

main.py

import yfinance as yf
import matplotlib.pyplot as plt
import matplotlib
from datetime import datetime, timedelta
import numpy as np
from sklearn.linear_model import LinearRegression
from sklearn.preprocessing import PolynomialFeatures
from sklearn.metrics import r2_score, mean_squared_error
from scipy import stats
import pandas as pd
from statsmodels.nonparametric.smoothers_lowess import lowess
import warnings
warnings.filterwarnings('ignore')

# 日本語フォントの設定
matplotlib.rcParams['font.family'] = 'MS Gothic'
matplotlib.rcParams['axes.unicode_minus'] = False  # マイナス記号の文字化け対策

def analyze_and_plot(symbol: str, pair_label: str):
    # 過去20年間のデータを取得
    end_date = datetime.now()
    start_date = end_date - timedelta(days=365*20)
    print(f"{pair_label}データをダウンロード中...")
    df = yf.download(symbol, start=start_date, end=end_date)
    y = df['Close'].values.flatten()
    X = np.arange(len(y)).reshape(-1, 1)
    n = len(y)

    # 将来10年分の予測
    future_days = 252 * 10
    X_future = np.arange(len(y) + future_days).reshape(-1, 1)
    future_dates = pd.date_range(start=df.index[-1], periods=future_days+1, freq='D')[1:]
    all_dates = df.index.append(future_dates)

    # LOWESS
    lowess_result = lowess(y, X.flatten(), frac=0.1, return_sorted=False)
    last_100_points = lowess_result[-100:]
    lowess_slope = (last_100_points[-1] - last_100_points[0]) / 100
    lowess_future = np.concatenate([
        lowess_result,
        [lowess_result[-1] + lowess_slope * i for i in range(1, future_days+1)]
    ])
    rmse_lowess = np.sqrt(mean_squared_error(y, lowess_result))
    r2_lowess = r2_score(y, lowess_result)

    # 線形回帰
    linear_model = LinearRegression()
    linear_model.fit(X, y)
    y_linear = linear_model.predict(X)
    y_linear_future = linear_model.predict(X_future)
    r2_linear = r2_score(y, y_linear)
    rmse_linear = np.sqrt(mean_squared_error(y, y_linear))
    df_res_linear = n - 2
    f_stat_linear = (r2_linear / 1) / ((1 - r2_linear) / df_res_linear)
    p_value_linear = 1 - stats.f.cdf(f_stat_linear, 1, df_res_linear)

    # 2次多項式
    poly2_features = PolynomialFeatures(degree=2)
    X_poly2 = poly2_features.fit_transform(X)
    X_poly2_future = poly2_features.fit_transform(X_future)
    poly2_model = LinearRegression()
    poly2_model.fit(X_poly2, y)
    y_poly2 = poly2_model.predict(X_poly2)
    y_poly2_future = poly2_model.predict(X_poly2_future)
    r2_poly2 = r2_score(y, y_poly2)
    rmse_poly2 = np.sqrt(mean_squared_error(y, y_poly2))
    df_res_poly2 = n - 3
    f_stat_poly2 = (r2_poly2 / 2) / ((1 - r2_poly2) / df_res_poly2)
    p_value_poly2 = 1 - stats.f.cdf(f_stat_poly2, 2, df_res_poly2)

    # 3次多項式
    poly3_features = PolynomialFeatures(degree=3)
    X_poly3 = poly3_features.fit_transform(X)
    X_poly3_future = poly3_features.fit_transform(X_future)
    poly3_model = LinearRegression()
    poly3_model.fit(X_poly3, y)
    y_poly3 = poly3_model.predict(X_poly3)
    y_poly3_future = poly3_model.predict(X_poly3_future)
    r2_poly3 = r2_score(y, y_poly3)
    rmse_poly3 = np.sqrt(mean_squared_error(y, y_poly3))
    df_res_poly3 = n - 4
    f_stat_poly3 = (r2_poly3 / 3) / ((1 - r2_poly3) / df_res_poly3)
    p_value_poly3 = 1 - stats.f.cdf(f_stat_poly3, 3, df_res_poly3)

    # グラフ
    plt.figure(figsize=(16, 9))
    plt.scatter(df.index, y, s=1.0, label='実際の価格', alpha=0.4, color='gray', zorder=1)
    plt.axvline(x=df.index[-1], color='black', linestyle=':', linewidth=1.2, alpha=0.4, label='予測開始点')
    plt.plot(all_dates, lowess_future, linewidth=2,
             label=f'★LOWESS (R²={r2_lowess:.4f}, RMSE={rmse_lowess:.2f})',
             linestyle='-', color='C0', alpha=0.9, zorder=5)
    plt.plot(all_dates, y_linear_future, linewidth=2,
             label=f'線形回帰 (R²={r2_linear:.4f}, RMSE={rmse_linear:.2f})',
             linestyle='--', color='C1', alpha=0.7, zorder=3)
    plt.plot(all_dates, y_poly2_future, linewidth=2,
             label=f'2次多項式回帰 (R²={r2_poly2:.4f}, RMSE={rmse_poly2:.2f})',
             linestyle='--', color='C2', alpha=0.7, zorder=3)
    plt.plot(all_dates, y_poly3_future, linewidth=2,
             label=f'3次多項式回帰 (R²={r2_poly3:.4f}, RMSE={rmse_poly3:.2f})',
             linestyle='--', color='C3', alpha=0.7, zorder=3)

    plt.title(f'{pair_label} 為替レート - 回帰分析と10年予測', fontsize=14, fontweight='bold')
    plt.xlabel('日付', fontsize=12)
    plt.ylabel('価格 (円)', fontsize=12)
    plt.legend(loc='best', fontsize=9)
    plt.grid(True, alpha=0.3)
    plt.tight_layout()
    print("グラフを表示中...")
    plt.show()

    # 結果出力
    print(f"\n=== {pair_label} 統計情報 ===")
    print(f"期間: {df.index[0].date()} ~ {df.index[-1].date()}")
    print(f"最高値: {df['Close'].max():.2f}円")
    print(f"最安値: {df['Close'].min():.2f}円")
    print(f"平均値: {df['Close'].mean():.2f}円")
    print(f"最新値: {df['Close'][-1]:.2f}円")

    print(f"\n=== {pair_label} より妥当性の高い分析結果 ===")
    print(f"★LOWESS: R²={r2_lowess:.4f}, RMSE={rmse_lowess:.2f}")

    print(f"\n=== {pair_label} 従来の回帰分析結果 ===")
    print(f"線形: R²={r2_linear:.4f}, RMSE={rmse_linear:.2f}, F={f_stat_linear:.4f}, p={p_value_linear:.2e}")
    print(f"2次: R²={r2_poly2:.4f}, RMSE={rmse_poly2:.2f}, F={f_stat_poly2:.4f}, p={p_value_poly2:.2e}")
    print(f"3次: R²={r2_poly3:.4f}, RMSE={rmse_poly3:.2f}, F={f_stat_poly3:.4f}, p={p_value_poly3:.2e}")

    print(f"\n=== {pair_label} 10年後の予測値 ===")
    print(f"LOWESS: {lowess_future[-1]:.2f}円")
    print(f"線形回帰: {y_linear_future[-1]:.2f}円")
    print(f"2次多項式回帰: {y_poly2_future[-1]:.2f}円")
    print(f"3次多項式回帰: {y_poly3_future[-1]:.2f}円")

def main():
    analyze_and_plot('USDJPY=X', 'USDJPY (ドル円)')

if __name__ == "__main__":
    main()