ランダムウォークへの挑戦㉑｜Q3の優位性は時間が変わっても残るのか（2026/8/29）

import os
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt


# ============================================================
# 設定
# ============================================================

INPUT_FILE = "results/trade_base.csv"

OUTPUT_DIR = "results"

os.makedirs(
    OUTPUT_DIR,
    exist_ok=True
)


OUTPUT_PERIOD_CSV = os.path.join(
    OUTPUT_DIR,
    "relative_band_period_analysis.csv"
)

OUTPUT_PERIOD_PF_CSV = os.path.join(
    OUTPUT_DIR,
    "relative_band_period_pf.csv"
)

OUTPUT_YEARLY_CSV = os.path.join(
    OUTPUT_DIR,
    "relative_band_yearly_analysis.csv"
)

OUTPUT_YEARLY_PF_CSV = os.path.join(
    OUTPUT_DIR,
    "relative_band_yearly_pf.csv"
)

OUTPUT_PERIOD_PF_GRAPH = os.path.join(
    OUTPUT_DIR,
    "relative_band_period_pf.png"
)

OUTPUT_CUMULATIVE_GRAPH = os.path.join(
    OUTPUT_DIR,
    "relative_band_cumulative_profit.png"
)

OUTPUT_Q3_PERIOD_GRAPH = os.path.join(
    OUTPUT_DIR,
    "q3_period_profit_factor.png"
)

OUTPUT_Q3_CUMULATIVE_GRAPH = os.path.join(
    OUTPUT_DIR,
    "q3_cumulative_profit.png"
)


# ============================================================
# データ読み込み
# ============================================================

df = pd.read_csv(
    INPUT_FILE
)


print(
    "========================================"
)

print(
    "RELATIVE BAND TIME ANALYSIS"
)

print(
    "========================================"
)

print(
    f"\n読み込み件数: {len(df)}"
)


# ============================================================
# 日時列の判定
# ============================================================

time_candidates = [
    "entry_time",
    "entry_datetime",
    "open_time",
    "EntryTime",
    "ENTRY_TIME"
]


time_column = None


for candidate in time_candidates:

    if candidate in df.columns:

        time_column = candidate

        break


if time_column is None:

    raise ValueError(
        "取引日時の列が見つかりません。"
        "entry_time等の列名を確認してください。"
    )


print(
    f"使用する日時列: {time_column}"
)


# ============================================================
# 必要列確認
# ============================================================

required_columns = [
    "entry_band_distance_pips",
    "entry_atr14_pips",
    "profit",
    time_column
]


missing_columns = [
    column
    for column in required_columns
    if column not in df.columns
]


if missing_columns:

    raise ValueError(
        "必要な列がありません: "
        + ", ".join(missing_columns)
    )


# ============================================================
# 数値変換
# ============================================================

for column in [
    "entry_band_distance_pips",
    "entry_atr14_pips",
    "profit"
]:

    df[column] = pd.to_numeric(
        df[column],
        errors="coerce"
    )


# ============================================================
# 日時変換
# ============================================================

df[time_column] = pd.to_datetime(
    df[time_column],
    errors="coerce"
)


# ============================================================
# 分析対象
# ============================================================

analysis_df = df[
    df[
        "entry_band_distance_pips"
    ].notna()
    &
    df[
        "entry_atr14_pips"
    ].notna()
    &
    df[
        "profit"
    ].notna()
    &
    df[
        time_column
    ].notna()
    &
    (
        df[
            "entry_atr14_pips"
        ]
        > 0
    )
].copy()


if analysis_df.empty:

    raise ValueError(
        "分析対象データがありません。"
    )


print(
    f"分析対象件数: {len(analysis_df)}"
)


# ============================================================
# Relative Band Distance
# ============================================================

analysis_df[
    "relative_band_distance"
] = (
    analysis_df[
        "entry_band_distance_pips"
    ]
    /
    analysis_df[
        "entry_atr14_pips"
    ]
)


analysis_df[
    "relative_band_distance"
] = (
    analysis_df[
        "relative_band_distance"
    ]
    .replace(
        [
            np.inf,
            -np.inf
        ],
        np.nan
    )
)


analysis_df = analysis_df[
    analysis_df[
        "relative_band_distance"
    ].notna()
].copy()


# ============================================================
# 時系列順に並び替え
# ============================================================

analysis_df = (
    analysis_df
    .sort_values(
        time_column
    )
    .reset_index(
        drop=True
    )
)


# ============================================================
# Relative Band Distance 四分位
# ============================================================

quartile_order = [
    "Q1",
    "Q2",
    "Q3",
    "Q4"
]


analysis_df[
    "relative_band_group"
] = pd.qcut(
    analysis_df[
        "relative_band_distance"
    ],
    q=4,
    labels=quartile_order
)


# ============================================================
# 年
# ============================================================

analysis_df[
    "year"
] = (
    analysis_df[
        time_column
    ]
    .dt
    .year
)


# ============================================================
# 取引順による4期間分割
# ============================================================

number_of_trades = len(
    analysis_df
)


analysis_df[
    "trade_order"
] = np.arange(
    number_of_trades
)


analysis_df[
    "time_period"
] = pd.qcut(
    analysis_df[
        "trade_order"
    ],
    q=4,
    labels=[
        "Period1",
        "Period2",
        "Period3",
        "Period4"
    ]
)


period_order = [
    "Period1",
    "Period2",
    "Period3",
    "Period4"
]


# ============================================================
# PF計算関数
# ============================================================

def calculate_pf(data):

    gross_profit = (
        data.loc[
            data[
                "profit"
            ]
            > 0,
            "profit"
        ]
        .sum()
    )


    gross_loss = abs(
        data.loc[
            data[
                "profit"
            ]
            < 0,
            "profit"
        ]
        .sum()
    )


    if gross_loss == 0:

        return np.nan


    return (
        gross_profit
        /
        gross_loss
    )


# ============================================================
# 成績計算関数
# ============================================================

def calculate_summary(data):

    trades = len(
        data
    )


    if trades == 0:

        return {
            "trades": 0,
            "wins": 0,
            "losses": 0,
            "win_rate": np.nan,
            "total_profit": np.nan,
            "average_profit": np.nan,
            "median_profit": np.nan,
            "profit_factor": np.nan
        }


    wins = (
        data[
            "profit"
        ]
        > 0
    ).sum()


    losses = (
        data[
            "profit"
        ]
        < 0
    ).sum()


    win_rate = (
        wins
        /
        trades
        * 100
    )


    return {
        "trades": trades,

        "wins": wins,

        "losses": losses,

        "win_rate": win_rate,

        "total_profit": (
            data[
                "profit"
            ]
            .sum()
        ),

        "average_profit": (
            data[
                "profit"
            ]
            .mean()
        ),

        "median_profit": (
            data[
                "profit"
            ]
            .median()
        ),

        "profit_factor": (
            calculate_pf(
                data
            )
        )
    }


# ============================================================
# 4期間 × Q1～Q4 分析
# ============================================================

period_rows = []


for period in period_order:

    for group in quartile_order:

        group_df = analysis_df[
            (
                analysis_df[
                    "time_period"
                ]
                == period
            )
            &
            (
                analysis_df[
                    "relative_band_group"
                ]
                == group
            )
        ].copy()


        stats = calculate_summary(
            group_df
        )


        period_rows.append(
            {
                "period": period,

                "group": group,

                **stats
            }
        )


period_df = pd.DataFrame(
    period_rows
)


# ============================================================
# 小数点整理
# ============================================================

period_df[
    "win_rate"
] = (
    period_df[
        "win_rate"
    ]
    .round(2)
)


for column in [
    "total_profit",
    "average_profit",
    "median_profit"
]:

    period_df[column] = (
        period_df[column]
        .round(2)
    )


period_df[
    "profit_factor"
] = (
    period_df[
        "profit_factor"
    ]
    .round(3)
)


print(
    "\n========================================"
)

print(
    "4期間 × Relative Band Distance"
)

print(
    "========================================"
)


print(
    period_df
)


period_df.to_csv(
    OUTPUT_PERIOD_CSV,
    index=False,
    encoding="utf-8-sig"
)


# ============================================================
# 期間別PF表
# ============================================================

period_pf = (
    period_df
    .pivot(
        index="period",
        columns="group",
        values="profit_factor"
    )
)


period_pf = period_pf.reindex(
    index=period_order,
    columns=quartile_order
)


print(
    "\n========================================"
)

print(
    "期間別PF"
)

print(
    "========================================"
)


print(
    period_pf
)


period_pf.to_csv(
    OUTPUT_PERIOD_PF_CSV,
    encoding="utf-8-sig"
)


# ============================================================
# 年別分析
# ============================================================

yearly_rows = []


years = sorted(
    analysis_df[
        "year"
    ]
    .dropna()
    .unique()
)


for year in years:

    for group in quartile_order:

        group_df = analysis_df[
            (
                analysis_df[
                    "year"
                ]
                == year
            )
            &
            (
                analysis_df[
                    "relative_band_group"
                ]
                == group
            )
        ].copy()


        stats = calculate_summary(
            group_df
        )


        yearly_rows.append(
            {
                "year": int(
                    year
                ),

                "group": group,

                **stats
            }
        )


yearly_df = pd.DataFrame(
    yearly_rows
)


yearly_df[
    "win_rate"
] = (
    yearly_df[
        "win_rate"
    ]
    .round(2)
)


for column in [
    "total_profit",
    "average_profit",
    "median_profit"
]:

    yearly_df[column] = (
        yearly_df[column]
        .round(2)
    )


yearly_df[
    "profit_factor"
] = (
    yearly_df[
        "profit_factor"
    ]
    .round(3)
)


print(
    "\n========================================"
)

print(
    "年別成績"
)

print(
    "========================================"
)


print(
    yearly_df
)


yearly_df.to_csv(
    OUTPUT_YEARLY_CSV,
    index=False,
    encoding="utf-8-sig"
)


# ============================================================
# 年別PF
# ============================================================

yearly_pf = (
    yearly_df
    .pivot(
        index="year",
        columns="group",
        values="profit_factor"
    )
)


yearly_pf = yearly_pf.reindex(
    columns=quartile_order
)


yearly_pf.to_csv(
    OUTPUT_YEARLY_PF_CSV,
    encoding="utf-8-sig"
)


# ============================================================
# 期間別PFグラフ
# ============================================================

fig, ax = plt.subplots(
    figsize=(10, 6)
)


period_pf.plot(
    kind="bar",
    ax=ax
)


ax.axhline(
    y=1.0,
    linestyle="--"
)


ax.set_xlabel(
    "Time Period"
)


ax.set_ylabel(
    "Profit Factor"
)


ax.set_title(
    "Profit Factor by Time Period and Relative Band Distance"
)


ax.tick_params(
    axis="x",
    rotation=0
)


ax.grid(
    axis="y",
    alpha=0.3
)


plt.tight_layout()


plt.savefig(
    OUTPUT_PERIOD_PF_GRAPH,
    dpi=150
)


plt.close()


# ============================================================
# Q3のみ期間別PF
# ============================================================

q3_period = (
    period_pf[
        "Q3"
    ]
)


fig, ax = plt.subplots(
    figsize=(8, 6)
)


q3_period.plot(
    kind="bar",
    ax=ax
)


ax.axhline(
    y=1.0,
    linestyle="--"
)


ax.set_xlabel(
    "Time Period"
)


ax.set_ylabel(
    "Profit Factor"
)


ax.set_title(
    "Q3 Profit Factor by Time Period"
)


ax.tick_params(
    axis="x",
    rotation=0
)


ax.grid(
    axis="y",
    alpha=0.3
)


plt.tight_layout()


plt.savefig(
    OUTPUT_Q3_PERIOD_GRAPH,
    dpi=150
)


plt.close()


# ============================================================
# Q1～Q4累積損益
# ============================================================

fig, ax = plt.subplots(
    figsize=(11, 7)
)


for group in quartile_order:

    group_df = analysis_df[
        analysis_df[
            "relative_band_group"
        ]
        == group
    ].copy()


    group_df = (
        group_df
        .sort_values(
            time_column
        )
    )


    group_df[
        "cumulative_profit"
    ] = (
        group_df[
            "profit"
        ]
        .cumsum()
    )


    ax.plot(
        group_df[
            time_column
        ],
        group_df[
            "cumulative_profit"
        ],
        label=group
    )


ax.axhline(
    y=0,
    linestyle="--"
)


ax.set_xlabel(
    "Time"
)


ax.set_ylabel(
    "Cumulative Profit"
)


ax.set_title(
    "Cumulative Profit by Relative Band Distance"
)


ax.legend()


ax.grid(
    alpha=0.3
)


plt.tight_layout()


plt.savefig(
    OUTPUT_CUMULATIVE_GRAPH,
    dpi=150
)


plt.close()


# ============================================================
# Q3のみ累積損益
# ============================================================

q3_df = analysis_df[
    analysis_df[
        "relative_band_group"
    ]
    == "Q3"
].copy()


q3_df = (
    q3_df
    .sort_values(
        time_column
    )
)


q3_df[
    "cumulative_profit"
] = (
    q3_df[
        "profit"
    ]
    .cumsum()
)


fig, ax = plt.subplots(
    figsize=(11, 7)
)


ax.plot(
    q3_df[
        time_column
    ],
    q3_df[
        "cumulative_profit"
    ]
)


ax.axhline(
    y=0,
    linestyle="--"
)


ax.set_xlabel(
    "Time"
)


ax.set_ylabel(
    "Cumulative Profit"
)


ax.set_title(
    "Q3 Cumulative Profit"
)


ax.grid(
    alpha=0.3
)


plt.tight_layout()


plt.savefig(
    OUTPUT_Q3_CUMULATIVE_GRAPH,
    dpi=150
)


plt.close()


# ============================================================
# Q3の期間安定性を表示
# ============================================================

print(
    "\n========================================"
)

print(
    "Q3 PERIOD STABILITY"
)

print(
    "========================================"
)


q3_stability = period_df[
    period_df[
        "group"
    ]
    == "Q3"
][
    [
        "period",
        "trades",
        "win_rate",
        "total_profit",
        "average_profit",
        "profit_factor"
    ]
]


print(
    q3_stability
)


# ============================================================
# Q3 PF > 1 の期間数
# ============================================================

q3_positive_periods = (
    q3_stability[
        "profit_factor"
    ]
    > 1
).sum()


print(
    "\nQ3 PF > 1 の期間数:",
    f"{q3_positive_periods} / 4"
)


# ============================================================
# Q3 PFの平均・標準偏差
# ============================================================

q3_pf_mean = (
    q3_stability[
        "profit_factor"
    ]
    .mean()
)


q3_pf_std = (
    q3_stability[
        "profit_factor"
    ]
    .std()
)


print(
    f"Q3 期間別PF平均: {q3_pf_mean:.3f}"
)


print(
    f"Q3 期間別PF標準偏差: {q3_pf_std:.3f}"
)


# ============================================================
# 出力ファイル
# ============================================================

print(
    "\n========================================"
)

print(
    "分析完了"
)

print(
    "========================================"
)


print(
    "\n出力ファイル:"
)


print(
    OUTPUT_PERIOD_CSV
)


print(
    OUTPUT_PERIOD_PF_CSV
)


print(
    OUTPUT_YEARLY_CSV
)


print(
    OUTPUT_YEARLY_PF_CSV
)


print(
    OUTPUT_PERIOD_PF_GRAPH
)


print(
    OUTPUT_CUMULATIVE_GRAPH
)


print(
    OUTPUT_Q3_PERIOD_GRAPH
)


print(
    OUTPUT_Q3_CUMULATIVE_GRAPH
)