Rewrite as a probabilistic model with walk-forward evaluation.

Replace the 2024 model (model.py, ~2000 lines) with the btcmodel package, the baseline for future work: - Forecasts are quantiles of log price at each horizon, scored with CRPS in a walk-forward backtest (origins every 30 days from 2014, horizons 1 month to 4 years). Skill is relative to a zero-drift random walk, with circular block-bootstrap intervals and a count of independent windows. - Development data stops at 2024-11-26, the last day the 2024 model saw. Later outcomes are a holdout, scored only by `backtest --holdout`. - Models: random_walk, drift_rw, and cycle (the 2024 model's cycle-position drift, now kernel-smoothed and recency-weighted). On development data nothing beats the random walk with confidence; cycle loses at every horizon. - Prices: the Investing.com archive moves to data/ (cut at 2024-11-26; its last row was intraday) and is extended with Coinbase daily closes by `update`. Also: Nix flake dev shell (Python 3.13, pandas 3), ruff in place of black, pytest suite, and a rewritten README. NOTES.md is removed as inaccurate, and poetry is dropped.
2026-09-24 02:19:02 -07:00
parent eefff47070
commit cfc27a38de
24 changed files with 1767 additions and 3124 deletions
@@ -0,0 +1,172 @@
+"""Charts. One colour per model, fixed by its position in MODELS."""
+
+from pathlib import Path
+
+import matplotlib
+
+matplotlib.use("Agg")
+
+import matplotlib.pyplot as plt  # noqa: E402
+import numpy as np  # noqa: E402
+import pandas as pd  # noqa: E402
+from matplotlib.ticker import FuncFormatter, LogLocator, PercentFormatter  # noqa: E402
+
+from .evaluate import COVERAGES, horizon_label  # noqa: E402
+from .forecast import Forecast  # noqa: E402
+from .halving import CYCLE_STARTS  # noqa: E402
+from .models import BASELINE, MODELS  # noqa: E402
+
+SURFACE = "#fcfcfb"
+INK = "#0b0b0b"
+INK_SECONDARY = "#52514e"
+MUTED = "#898781"
+GRID = "#e1e0d9"
+AXIS = "#c3c2b7"
+SERIES = ["#2a78d6", "#eb6834", "#1baf7a", "#eda100", "#e87ba4", "#008300", "#4a3aa7", "#e34948"]
+
+plt.rcParams.update(
+    {
+        "figure.facecolor": SURFACE,
+        "axes.facecolor": SURFACE,
+        "savefig.facecolor": SURFACE,
+        "axes.edgecolor": AXIS,
+        "axes.linewidth": 0.8,
+        "axes.labelcolor": INK_SECONDARY,
+        "axes.titlecolor": INK,
+        "axes.titlesize": 11,
+        "axes.titleweight": "semibold",
+        "axes.titlelocation": "left",
+        "axes.spines.top": False,
+        "axes.spines.right": False,
+        "axes.grid": True,
+        "grid.color": GRID,
+        "grid.linewidth": 0.8,
+        "xtick.color": MUTED,
+        "ytick.color": MUTED,
+        "xtick.labelcolor": INK_SECONDARY,
+        "ytick.labelcolor": INK_SECONDARY,
+        "legend.frameon": False,
+        "legend.labelcolor": INK_SECONDARY,
+        "lines.linewidth": 2,
+        "lines.solid_capstyle": "round",
+        "lines.solid_joinstyle": "round",
+        "font.family": "sans-serif",
+        "font.size": 9,
+    }
+)
+
+
+def model_color(name: str) -> str:
+    return SERIES[list(MODELS).index(name) % len(SERIES)]
+
+
+def price_formatter(x, _=None) -> str:
+    for scale, suffix in ((1e9, "B"), (1e6, "M"), (1e3, "K")):
+        if x >= scale:
+            return f"${x / scale:.3g}{suffix}"
+    return f"${x:.3g}"
+
+
+def fan_chart(history: pd.DataFrame, forecasts: dict[str, Forecast], path: Path) -> None:
+    """History plus 50/80/95% intervals, one panel per model on shared axes."""
+    fig, axes = plt.subplots(
+        len(forecasts), 1, figsize=(10, 3.2 * len(forecasts)), sharex=True, sharey=True
+    )
+    axes = np.atleast_1d(axes)
+    shown = history[history.index >= history.index[-1] - pd.Timedelta(days=6 * 365)]
+    for ax, (name, f) in zip(axes, forecasts.items(), strict=True):
+        color = model_color(name)
+        for c in sorted(COVERAGES, reverse=True):
+            lo, hi = f.interval(c)
+            ax.fill_between(f.dates, np.exp(lo), np.exp(hi), color=color, alpha=0.1, lw=0)
+        median = np.exp(f.quantile(0.5))
+        ax.plot(shown.index, shown["close"], color=INK, lw=1.2)
+        ax.plot(f.dates, median, color=color)
+        ax.annotate(
+            f"median {price_formatter(median[-1])}",
+            (f.dates[-1], median[-1]),
+            xytext=(6, 0),
+            textcoords="offset points",
+            va="center",
+            color=INK_SECONDARY,
+        )
+        for start in CYCLE_STARTS:
+            if shown.index[0] <= start <= f.dates[-1]:
+                ax.axvline(start, color=AXIS, lw=0.8, zorder=0)
+        ax.set_yscale("log")
+        ax.yaxis.set_major_locator(LogLocator(base=10, subs=(1, 2, 5)))
+        ax.yaxis.set_major_formatter(FuncFormatter(price_formatter))
+        ax.yaxis.set_minor_formatter(FuncFormatter(lambda *_: ""))
+        ax.set_title(f"{name}: forecast from {f.origin:%Y-%m-%d}")
+    axes[0].fill_between([], [], color=MUTED, alpha=0.3, label="50% interval")
+    axes[0].fill_between([], [], color=MUTED, alpha=0.2, label="80% interval")
+    axes[0].fill_between([], [], color=MUTED, alpha=0.1, label="95% interval")
+    axes[0].legend(loc="upper left")
+    axes[-1].set_xlabel("vertical lines: halvings (future ones projected)", color=MUTED)
+    fig.tight_layout()
+    fig.savefig(path, dpi=150)
+    plt.close(fig)
+
+
+def skill_chart(summary: pd.DataFrame, path: Path, title: str) -> None:
+    """CRPS skill vs the random walk, by horizon, with bootstrap intervals."""
+    horizons = sorted(summary["horizon"].unique())
+    x = np.arange(len(horizons))
+    fig, ax = plt.subplots(figsize=(8, 4.5))
+    ax.axhline(0, color=model_color(BASELINE), lw=2, label=BASELINE)
+    for name, g in summary[summary.model != BASELINE].groupby("model", sort=False):
+        g = g.set_index("horizon").reindex(horizons)
+        color = model_color(name)
+        ax.fill_between(x, g.skill_lo, g.skill_hi, color=color, alpha=0.1, lw=0)
+        ax.plot(x, g.skill, color=color, marker="o", ms=6, mec=SURFACE, mew=2, label=name)
+        ax.annotate(
+            name,
+            (x[-1], g.skill.iloc[-1]),
+            xytext=(8, 0),
+            textcoords="offset points",
+            va="center",
+            color=INK_SECONDARY,
+        )
+    ax.set_xticks(x, [horizon_label(h) for h in horizons])
+    ax.set_xlabel("forecast horizon")
+    ax.set_ylabel("CRPS skill vs random walk (higher is better)")
+    ax.yaxis.set_major_formatter(PercentFormatter(1.0, decimals=0))
+    ax.set_title(title)
+    ax.legend(loc="lower left")
+    fig.tight_layout()
+    fig.savefig(path, dpi=150)
+    plt.close(fig)
+
+
+def calibration_chart(summary: pd.DataFrame, path: Path, title: str) -> None:
+    """How often each nominal interval contained the outcome, by horizon."""
+    horizons = sorted(summary["horizon"].unique())
+    x = np.arange(len(horizons))
+    fig, axes = plt.subplots(1, len(COVERAGES), figsize=(12, 4), sharey=True)
+    for ax, c in zip(axes, COVERAGES, strict=True):
+        ax.axhline(c, color=INK_SECONDARY, lw=1)
+        ax.annotate(
+            "target", (x[-1], c), xytext=(0, 4), textcoords="offset points", ha="right", color=MUTED
+        )
+        for name, g in summary.groupby("model", sort=False):
+            g = g.set_index("horizon").reindex(horizons)
+            ax.plot(
+                x,
+                g[f"cov{c:.0%}"],
+                color=model_color(name),
+                marker="o",
+                ms=6,
+                mec=SURFACE,
+                mew=2,
+                label=name,
+            )
+        ax.set_xticks(x, [horizon_label(h) for h in horizons])
+        ax.set_title(f"{c:.0%} interval")
+        ax.set_ylim(0, 1.02)
+        ax.yaxis.set_major_formatter(PercentFormatter(1.0, decimals=0))
+    axes[0].set_ylabel("share of outcomes inside")
+    axes[-1].legend(loc="lower left")
+    fig.suptitle(title, x=0.01, ha="left", color=INK, fontsize=11, fontweight="semibold")
+    fig.tight_layout()
+    fig.savefig(path, dpi=150)
+    plt.close(fig)