import pandas as pd import plotly.graph_objects as go from data.schema import Out # Realised P&L is context, so it stays neutral; the forecast # lines share one hue; exceedances use the critical status # color plus an x marker and a legend label. PNL_COLOR = "#b4b2a9" VAR_COLOR = "#2a78d6" ES_COLOR = "#184f95" EXCEEDANCE_COLOR = "#d03b3b" EXPECTED_COLOR = "#52514e" def _select(forecasts, confidence): selected = forecasts[ forecasts[Out.CONFIDENCE] == confidence ] if selected.empty: raise ValueError( f"No backtest results at confidence {confidence}." ) return selected.sort_values(Out.DATE) def _break_at_gaps(df, gap_days): """ Insert an empty row after every gap longer than `gap_days` so lines are not drawn across periods without data. """ if not gap_days: return df gaps = df[Out.DATE].diff() > pd.Timedelta(days=gap_days) breaks = pd.DataFrame( {Out.DATE: df.loc[gaps, Out.DATE] - pd.Timedelta(days=1)} ) return ( pd.concat([df, breaks]) .sort_values(Out.DATE) .reset_index(drop=True) ) def plot_backtest_pnl( forecasts, confidence, gap_days=None, ): """ Realised P&L against the forecast -VaR and -ES, with exceedances marked. """ df = _select(forecasts, confidence) exceeded = df[df[Out.EXCEEDANCE]] df = _break_at_gaps(df, gap_days) fig = go.Figure() fig.add_trace( go.Bar( x=df[Out.DATE], y=df[Out.REALIZED_PNL], name="Realized P&L", marker_color=PNL_COLOR, hovertemplate="%{y:,.0f}Realized P&L", ) ) fig.add_trace( go.Scatter( x=df[Out.DATE], y=-df[Out.VAR], mode="lines", name=f"-VaR {confidence:.1%}", line=dict(color=VAR_COLOR, width=2), hovertemplate="%{y:,.0f}-VaR", ) ) fig.add_trace( go.Scatter( x=df[Out.DATE], y=-df[Out.ES], mode="lines", name=f"-ES {confidence:.1%}", line=dict(color=ES_COLOR, width=2, dash="dot"), hovertemplate="%{y:,.0f}-ES", ) ) fig.add_trace( go.Scatter( x=exceeded[Out.DATE], y=exceeded[Out.REALIZED_PNL], mode="markers", name=f"Exceedance ({len(exceeded)})", marker=dict( color=EXCEEDANCE_COLOR, symbol="x", size=11, line=dict(width=1, color=EXCEEDANCE_COLOR), ), hovertemplate="%{y:,.0f}Exceedance", ) ) fig.update_layout( title=( f"Out-of-sample backtest: realized P&L vs " f"forecast VaR ({confidence:.1%})" ), xaxis_title="Date", yaxis_title="P&L", template="plotly_white", hovermode="x unified", bargap=0.3, ) return fig def plot_cumulative_exceedances( forecasts, confidence, gap_days=None, ): """ Observed exceedances over time against the number a correctly calibrated model would produce. """ df = _select(forecasts, confidence).reset_index(drop=True) df[Out.EXCEEDANCES] = df[Out.EXCEEDANCE].astype(int).cumsum() df[Out.EXPECTED_EXCEEDANCES] = (1 - confidence) * (df.index + 1) df = _break_at_gaps(df, gap_days) observed = df[Out.EXCEEDANCES] expected = df[Out.EXPECTED_EXCEEDANCES] fig = go.Figure() fig.add_trace( go.Scatter( x=df[Out.DATE], y=expected, mode="lines", name="Expected", line=dict(color=EXPECTED_COLOR, width=2, dash="dash"), hovertemplate="%{y:.2f}Expected", ) ) fig.add_trace( go.Scatter( x=df[Out.DATE], y=observed, mode="lines", name="Observed", line=dict(color=VAR_COLOR, width=2, shape="hv"), hovertemplate="%{y}Observed", ) ) fig.update_layout( title=f"Cumulative exceedances ({confidence:.1%})", xaxis_title="Date", yaxis_title="Exceedances", template="plotly_white", hovermode="x unified", ) return fig