# app.py — Macro Cycle Composite (2 sections, production style)

import math
from datetime import datetime, date, timedelta

import numpy as np
import pandas as pd
import pandas_datareader.data as web
import yfinance as yf

import streamlit as st
import plotly.graph_objects as go
from plotly.subplots import make_subplots
import plotly.express as px

# --------------------------- Page config ---------------------------
st.set_page_config(
    page_title="Market Cycle Composite",
    layout="wide",
    initial_sidebar_state="expanded"
)

st.title("Market Cycle Phases")

st.markdown(
    "Classifies the market cycle each period: **Early**, **Mid-Late**, **Decline**, or **Uncertain**. "
    "The solid line is the composite score; the dashed line is S&P 500 YoY. "
    "Colored bands show the active phase over time. "
    "Use the sidebar to set the sample start, smoothing, slope lookback, thresholds, and minimum run. "
    "Click **Run Analysis** to update the charts."
)

# --------------------------- Constants -----------------------------
TODAY_PLUS_1 = (date.today() + timedelta(days=1))
DEFAULT_START = date(2000, 1, 1)

FRED_MAP = {
    'LEI'               : 'USSLIND',
    'Philly Manuf Diff' : 'GACDFSA066MSFRBPHI',
    'Texas Serv Diff'   : 'TSSOSBACTUAMFRBDAL',
    'Capacity Util'     : 'CUMFNS',
    'BBK Leading'       : 'BBKMLEIX',
    'CFNAI 3MMA'        : 'CFNAIMA3',
    'Core CPI'          : 'CPILFESL',
    'Core PCE'          : 'PCEPILFE',
    'Hourly Wage'       : 'CES0500000003',
    'PPI'               : 'PPIACO',
    'Commodities'       : 'PALLFNFINDEXM',
    '10Y'               : 'DGS10',
    'HY OAS'            : 'BAMLH0A0HYM2',
    'StLouis FSI'       : 'STLFSI4',
}

PHASE_COLORS = {
    'Early'     : '#54d62c',
    'Mid-Late'  : '#3fa1ff',
    'Decline'   : '#ff4c4c',
    'Uncertain' : '#ffd400'
}

BG = "#0e1117"

# --------------------------- Caching -------------------------------
@st.cache_data(show_spinner=False)
def load_macro_series(start: date, end: date) -> pd.DataFrame:
    raw = {}
    for k, tkr in FRED_MAP.items():
        s = web.DataReader(tkr, 'fred', start, end).squeeze()
        raw[k] = s
    df = pd.DataFrame(raw)
    return df

@st.cache_data(show_spinner=False)
def load_spx(start: date, end: date) -> pd.Series:
    data = yf.download('^GSPC', start=start, end=end, auto_adjust=False, progress=False)
    data.columns = data.columns.get_level_values(0)
    spx = data['Adj Close'].resample('ME').last().ffill()
    return spx

# --------------------------- Sidebar -------------------------------
with st.sidebar:
    st.title("Parameters")

    with st.expander("Data Range", expanded=False):
        start_date = st.date_input(
            "Start Date",
            value=DEFAULT_START,
            min_value=date(1950, 1, 1),
            max_value=TODAY_PLUS_1,
            help=(
                "Start date for all series. "
                "Earlier = more history, heavier load. "
                "Later = less history."
            )
        )
        st.caption(f"End Date: {TODAY_PLUS_1.isoformat()} (fixed to today + 1)")

        freq = st.selectbox(
            "Resample Frequency",
            options=["ME", "QE"],
            index=0,
            help=(
                "Aggregation for all series. "
                "ME = month end. QE = quarter end. "
                "Higher aggregation smooths high-frequency noise."
            )
        )

    with st.expander("Composite & Phase Parameters", expanded=False):
        smooth_window = st.number_input(
            "Composite Smoothing Window (months)",
            value=2, min_value=1, max_value=12, step=1,
            help=(
                "Moving average on the composite. "
                "Increase to smooth more (slower). "
                "Decrease to react faster."
            )
        )
        slope_window = st.number_input(
            "Slope Lookback (months)",
            value=3, min_value=1, max_value=12, step=1,
            help=(
                "Difference span used for slope. "
                "Higher = coarser trend. "
                "Lower = more sensitive."
            )
        )
        comp_thr = st.number_input(
            "Composite Threshold",
            value=0.15, step=0.05, format="%.2f",
            help=(
                "Band around zero for phase changes. "
                "Increase = fewer flips. "
                "Decrease = earlier shifts."
            )
        )
        slope_thr = st.number_input(
            "Slope Threshold",
            value=0.005, step=0.001, format="%.3f",
            help=(
                "Minimum slope for Early/Decline. "
                "Increase to demand stronger momentum. "
                "Decrease to catch weaker turns."
            )
        )
        min_run = st.number_input(
            "Minimum Phase Run (months)",
            value=6, min_value=1, max_value=24, step=1,
            help=(
                "Median/mode filter on phases. "
                "Higher enforces longer runs. "
                "Lower allows quicker switches."
            )
        )

    run = st.button("Run Analysis", help="Compute the composite and render both sections.")

# --------------------------- Run pipeline --------------------------
if run:
    prog = st.progress(0, text="Loading data...")
    try:
        # Load & resample
        with st.spinner("Loading macro series..."):
            df_raw_daily = load_macro_series(start_date, TODAY_PLUS_1)
        prog.progress(20, text="Resampling...")

        df = df_raw_daily.resample(freq).last().ffill()

        # ====== TRANSFORMS & COMPOSITE ======
        yoy     = lambda s: s.pct_change(12) * 100
        delta12 = lambda s: s.diff(12)
        invert  = lambda s: -s

        tr = df.copy()
        tr['LEI']           = yoy(tr['LEI'])
        tr['Capacity Util'] = yoy(tr['Capacity Util'])
        tr['BBK Leading']   = yoy(tr['BBK Leading'])
        for c in ['Core CPI','Core PCE','Hourly Wage','PPI','Commodities']:
            tr[c] = yoy(tr[c])
        tr['10Y']         = invert(delta12(tr['10Y']))
        tr['HY OAS']      = invert(delta12(tr['HY OAS']))
        tr['StLouis FSI'] = invert(tr['StLouis FSI'])

        infl_cols = ['Core CPI','Core PCE','Hourly Wage','PPI','Commodities']
        tr['Inflation'] = tr[infl_cols].mean(axis=1)

        inputs = [
            'LEI','Philly Manuf Diff','Texas Serv Diff',
            'Capacity Util','BBK Leading','CFNAI 3MMA',
            'Inflation','10Y','HY OAS','StLouis FSI'
        ]

        z = tr[inputs].apply(lambda s: (s - s.mean())/s.std())
        comp = z.mean(axis=1)
        comp_sm = comp.rolling(int(smooth_window), min_periods=1).mean()
        slope = comp.diff(int(slope_window))

        # Phase classification
        cond_early   = (comp < -comp_thr) & (slope >  slope_thr)
        cond_midlate =  comp >  comp_thr
        cond_decline = (comp < -comp_thr) & (slope < -slope_thr)
        cond_unc     = comp.abs() <= comp_thr

        phase = pd.Series('Mid-Late', index=comp.index)
        phase[cond_early]   = 'Early'
        phase[cond_decline] = 'Decline'
        phase[cond_unc]     = 'Uncertain'

        code_map = {'Early':0, 'Mid-Late':1, 'Decline':2, 'Uncertain':3}
        inv_map  = {v:k for k,v in code_map.items()}
        codes = phase.map(code_map)

        smoothed_codes = codes.rolling(
            window=int(min_run), center=True, min_periods=1
        ).apply(lambda x: pd.Series(x).value_counts().idxmax(), raw=False)

        phase = smoothed_codes.round().astype(int).map(inv_map)

        prog.progress(55, text="Loading equity series...")
        spx = load_spx(start_date, TODAY_PLUS_1)
        spx_yoy = spx.pct_change(12) * 100
        if freq != 'ME':
            spx_yoy = spx_yoy.resample(freq).last().ffill()

        # ================== SECTION 1 — COMPOSITE ==================
        st.header("Market Cycle Composite Indicator")

        with st.expander("Methodology", expanded=False):
            st.markdown("#### What you’re looking at")
            st.write(
                "One score that summarizes many macro and market series. "
                "We also show the S&P 500 YoY for context. "
                "Colored bands mark the phase we infer at each date."
            )

            st.markdown("#### How we build the score")
            st.write("1) Put all series on the same timeline.")
            st.write("We keep the last value each period and fill gaps with the last known value.")
            st.latex(r"X^{(F)}_{t} = X_{\tau(t)},\;\; \tau(t)=\text{last timestamp in period }t")
            st.latex(r"\tilde{X}_t = \begin{cases} X_t, & \text{if observed}\\ \tilde{X}_{t-1}, & \text{if missing}\end{cases}")

            st.write("2) Convert each series to a comparable change.")
            st.write("Most series use year-over-year percent change:")
            st.latex(r"\mathrm{YoY}(X_t)=100\left(\frac{X_t}{X_{t-12}}-1\right)")
            st.write("Rates and spreads use 12-month differences (and we flip the sign when higher is worse):")
            st.latex(r"\Delta_{12}(X_t)=X_t-X_{t-12},\qquad \tilde{X}_t=-\Delta_{12}(X_t)")
            st.write("Stress metrics are inverted so higher stress lowers the score:")
            st.latex(r"\tilde{X}_t=-X_t")

            st.write("3) Build an inflation block from five series and average them:")
            st.latex(r"\mathrm{Inflation}_t=\frac{1}{5}\sum_{i=1}^{5}X_{i,t}")

            st.write("4) Standardize each input so all have the same scale.")
            st.latex(r"Z_{i,t}=\frac{X_{i,t}-\mu_i}{\sigma_i}")

            st.write("5) Average the standardized inputs to get the composite.")
            st.latex(r"C_t=\frac{1}{N}\sum_{i=1}^{N}Z_{i,t}")

            st.write("6) Smooth the composite to cut noise.")
            st.latex(r"\bar{C}_t=\frac{1}{W}\sum_{k=0}^{W-1}C_{t-k}")

            st.write("7) Measure recent direction with a simple slope.")
            st.latex(r"S_t=C_t-C_{t-W_s}")

            st.write("8) Assign a phase using two thresholds.")
            st.latex(r"\text{Early: } C_t<-\theta_c \ \wedge\ S_t>+\theta_s")
            st.latex(r"\text{Decline: } C_t<-\theta_c \ \wedge\ S_t<-\theta_s")
            st.latex(r"\text{Mid\text{-}Late: } C_t>+\theta_c")
            st.latex(r"\text{Uncertain: } |C_t|\le\theta_c")

            st.write("9) Stabilize phases with a centered majority vote over a short window.")
            st.latex(r"\hat{P}_t=\operatorname{mode}\{P_{t-k},\ldots,P_{t+k}\},\;\; m=2k+1")

            st.markdown("#### How the sidebar settings change results")
            st.write("- **Resample Frequency**: Month-end or quarter-end. Higher aggregation is smoother.")
            st.write("- **Composite Smoothing Window**: Larger = smoother, slower to react.")
            st.write("- **Slope Lookback**: Larger = slower slope, fewer flips.")
            st.write("- **Composite Threshold**: Larger = fewer phase changes.")
            st.write("- **Slope Threshold**: Larger = only strong turns count as Early/Decline.")
            st.write("- **Minimum Phase Run**: Larger = longer required runs, fewer whipsaws.")

        # Composite plot
        try:
            fig_comp = make_subplots(specs=[[{"secondary_y": True}]])
            fig_comp.update_layout(
                template="plotly_dark",
                height=520,
                margin=dict(l=60, r=20, t=60, b=40),
                title_text="Market-Cycle Composite Indicator",
                xaxis_rangeslider_visible=False,
                paper_bgcolor=BG,          # <— page background of the figure
                plot_bgcolor=BG,           # <— plotting area background
                font=dict(color="white"),
                legend=dict(
                    bgcolor="rgba(14,17,23,0)",  # transparent over the same bg tone
                    font=dict(color="white"),
                    title_font=dict(color="white")
                )
            )

            # Phase shading
            mask = phase.copy()
            grp = (mask != mask.shift()).cumsum()
            for ph, color in PHASE_COLORS.items():
                for _, span in mask[mask == ph].groupby(grp):
                    x0 = pd.Timestamp(span.index[0]).to_pydatetime()
                    x1 = pd.Timestamp(span.index[-1]).to_pydatetime()
                    fig_comp.add_shape(
                        type="rect",
                        x0=x0, x1=x1, y0=0, y1=1,
                        xref="x1", yref="paper",
                        fillcolor=color, opacity=0.22,
                        layer="below", line_width=0
                    )

            # Composite and zero
            fig_comp.add_trace(
                go.Scatter(
                    x=comp_sm.index, y=comp_sm, mode='lines',
                    line=dict(width=2), name=f'Cycle Composite ({int(smooth_window)}m MA)'
                ),
                secondary_y=False
            )
            fig_comp.add_hline(y=0, line_color='rgba(255,255,255,0.6)', line_width=1)

            # SPX YoY on secondary axis
            fig_comp.add_trace(
                go.Scatter(
                    x=spx_yoy.index, y=spx_yoy,
                    mode='lines', line=dict(width=2, dash='dash', color='#00d084'),
                    name='S&P 500 YoY %'
                ),
                secondary_y=True
            )

            # Legend keys for phases
            for ph, color in PHASE_COLORS.items():
                fig_comp.add_trace(
                    go.Scatter(
                        x=[None], y=[None], mode='lines',
                        line=dict(color=color, width=10),
                        name=ph, showlegend=True
                    ),
                    secondary_y=False
                )

            # Axes styling
            fig_comp.update_xaxes(
                tickformat='%Y',
                dtick="M12",
                title_font=dict(color="white"),
                tickfont=dict(color="white"),
                tickcolor="white",
                gridcolor="rgba(255,255,255,0.10)",
                zerolinecolor="rgba(255,255,255,0.15)",
                linecolor="rgba(255,255,255,0.15)",
                ticks="outside"
            )
            fig_comp.update_yaxes(
                title_text='Composite Z-Score',
                secondary_y=False,
                title_font=dict(color="white"),
                tickfont=dict(color="white"),
                tickcolor="white",
                gridcolor="rgba(255,255,255,0.10)",
                zerolinecolor="rgba(255,255,255,0.15)",
                linecolor="rgba(255,255,255,0.15)",
                ticks="outside"
            )
            fig_comp.update_yaxes(
                title_text='S&P 500 YoY %',
                secondary_y=True,
                range=[-80, 80],
                title_font=dict(color="white"),
                tickfont=dict(color="white"),
                tickcolor="white"
            )

            st.plotly_chart(fig_comp, use_container_width=True)
        except Exception:
            st.error("Failed to render the composite chart.")

        prog.progress(85, text="Computing interpretation...")

        # Dynamic Interpretation (richer, more explanatory)
        try:
            phase_series    = pd.Series(phase, index=comp.index)
            current_date    = phase_series.index[-1]
            current_phase   = phase_series.iloc[-1]
            current_comp    = comp_sm.loc[current_date]
            current_comp_raw = comp.loc[current_date]
            current_slope   = slope.loc[current_date]
            current_spx_yoy = spx_yoy.loc[current_date]

            # Percentiles (rank-based, full sample)
            comp_pct = float(comp_sm.dropna().rank(pct=True).loc[current_date] * 100)
            spx_pct  = float(spx_yoy.dropna().rank(pct=True).loc[current_date] * 100)

            # Phase run length (periods and months)
            changes = phase_series.ne(phase_series.shift())
            last_change_idx = changes[changes].index[-1]
            periods_in_phase = phase_series.index.get_loc(current_date) - phase_series.index.get_loc(last_change_idx) + 1
            if freq == "QE":
                months_in_phase = periods_in_phase * 3
                period_label = "quarters"
            else:
                months_in_phase = periods_in_phase
                period_label = "months"

            # Breadth and contributors
            z_now = z.loc[current_date].dropna()
            inputs_now = z_now.reindex(inputs).dropna()
            breadth_pos = float((inputs_now > 0).mean() * 100)
            top_pos = inputs_now.sort_values(ascending=False).head(3)
            top_neg = inputs_now.sort_values(ascending=True).head(3)

            def fmt_contrib(s):
                return ", ".join([f"{k} ({v:+.1f}σ)" for k, v in s.items()]) if len(s) else "n/a"

            # Phase-specific read-through and triggers
            phase_notes = []
            triggers = []

            if current_phase == 'Early':
                phase_notes += [
                    "Growth is turning up from a weak base.",
                    "Leading activity improves. Credit spreads narrow.",
                    "Rate momentum eases on a 12-month view."
                ]
                triggers += [
                    f"Upside: composite raw > {comp_thr:.2f}.",
                    f"Risk: slope < -{slope_thr:.3f} or composite raw > {-comp_thr:.2f}."
                ]
            elif current_phase == 'Mid-Late':
                phase_notes += [
                    "Growth remains above average but is slowing.",
                    "Pricing pressure and policy tightness rise.",
                    "Quality bias tends to help risk control."
                ]
                triggers += [
                    f"Loss of momentum: composite raw < {comp_thr:.2f}.",
                    f"Downside break: composite raw < {-comp_thr:.2f}."
                ]
            elif current_phase == 'Decline':
                phase_notes += [
                    "Activity contracts. Risk appetite weakens.",
                    "Credit and liquidity conditions worsen.",
                    "Drawdown risk is elevated relative to trend."
                ]
                triggers += [
                    f"Repair: slope > +{slope_thr:.3f}.",
                    f"Exit contraction: composite raw > {-comp_thr:.2f}."
                ]
            elif current_phase == 'Uncertain':
                phase_notes += [
                    "Signals conflict. Noise is high.",
                    "Avoid strong tilts until direction clears.",
                    "Use position sizing and stops."
                ]
                triggers += [
                    f"Upside regime: composite raw > {comp_thr:.2f}.",
                    f"Early setup: composite raw < {-comp_thr:.2f} and slope > +{slope_thr:.3f}.",
                    f"Decline setup: composite raw < {-comp_thr:.2f} and slope < -{slope_thr:.3f}."
                ]
            else:
                phase_notes.append("Phase classification unavailable.")

            # Build markdown
            interp_md = f"""
**As of {current_date.date()}**

- Phase: **{current_phase}** for {periods_in_phase} {period_label} (~{months_in_phase} months).
- Composite (smoothed): {current_comp:.2f} (p{comp_pct:.0f}). Raw: {current_comp_raw:.2f}.
- Slope over {int(slope_window)}m: {current_slope:+.3f}.
- S&P 500 YoY: {current_spx_yoy:.1f}% (p{spx_pct:.0f}).
- Breadth: {breadth_pos:.0f}% of inputs > 0.

**What drives the score now**
- Positive: {fmt_contrib(top_pos)}
- Negative: {fmt_contrib(top_neg)}

**Phase read-through**
""" + "\n".join([f"- {line}" for line in phase_notes]) + """

**Triggers to watch**
""" + "\n".join([f"- {t}" for t in triggers])

            with st.expander("Dynamic Interpretation", expanded=False):
                st.markdown(interp_md)
        except Exception:
            st.error("Failed to produce the interpretation.")

        # ================== SECTION 2 — RAW SERIES ==================
        prog.progress(95, text="Rendering input grid...")

        st.header("Macro Input Series (Resampled)")

        with st.expander("Methodology", expanded=False):
            st.write("Resample each series to the selected frequency (period end).")
            st.latex(r"X^{(F)}_{t} = X_{\tau(t)} \quad \text{with } \tau(t) = \text{last timestamp in period } t")
            st.write("Forward-fill missing observations to avoid gaps in aligned panels.")
            st.latex(r"\tilde{X}_t = \begin{cases} X_t, & \text{if observed} \\ \tilde{X}_{t-1}, & \text{otherwise} \end{cases}")
            st.write("No transforms are applied in this section. It is a clean view of inputs after resampling.")

        try:
            df_view = df.copy()
            n_series = len(df_view.columns)
            ncols = 3
            nrows = math.ceil(n_series / ncols)

            fig_grid = make_subplots(
                rows=nrows, cols=ncols,
                shared_xaxes=False,
                subplot_titles=list(df_view.columns)
            )

            xmin, xmax = df_view.index.min(), df_view.index.max()
            for i, col in enumerate(df_view.columns, start=1):
                r = (i-1)//ncols + 1
                c = (i-1)%ncols + 1
                fig_grid.add_trace(
                    go.Scatter(
                        x=df_view.index, y=df_view[col],
                        mode='lines',
                        line=dict(width=1.5),
                        name=col, showlegend=False
                    ),
                    row=r, col=c
                )

            for i in range(1, nrows*ncols + 1):
                xaxis_key = f'xaxis{i}' if i > 1 else 'xaxis'
                if xaxis_key in fig_grid.layout:
                    fig_grid.layout[xaxis_key].update(
                        range=[xmin, xmax],
                        tickformat='%Y',
                        tickfont=dict(color="white"),
                        tickcolor="white"
                    )
            
            fig_grid.update_layout(
                template="plotly_dark",
                height=max(360, nrows*260),
                title_text="All Input Series",
                margin=dict(l=40, r=10, t=50, b=40),
                paper_bgcolor=BG,   # <— match Streamlit dark bg
                plot_bgcolor=BG,    # <— match Streamlit dark bg
                font=dict(color="white")
            )

            for i in range(1, nrows*ncols + 1):
                yaxis_key = f'yaxis{i}' if i > 1 else 'yaxis'
                if yaxis_key in fig_grid.layout:
                    fig_grid.layout[yaxis_key].update(
                        tickfont=dict(color="white"),
                        tickcolor="white"
                    )

            st.plotly_chart(fig_grid, use_container_width=True)
        except Exception:
            st.error("Failed to render the input grid.")

        prog.progress(100, text="Done.")
    except Exception:
        st.error("Processing failed. Adjust parameters and try again.")
else:
    st.info("Set parameters in the sidebar and click **Run Analysis**.")


# Hide default Streamlit style
st.markdown(
    """
    <style>
    #MainMenu {visibility: hidden;}
    footer {visibility: hidden;}
    </style>
    """,
    unsafe_allow_html=True
)