Create tarefa3

tarefa3

@@ -0,0 +1,131 @@
+# ============================================================
+# VAR-VECM • IRF • FEVD  (Séries: PIB_real, Core PCE, Petróleo WTI)
+# Requer: streamlit ≥ 1.25, pandas ≥ 2.0, statsmodels ≥ 0.14
+# ============================================================
+
+import streamlit as st
+import pandas as pd
+import numpy as np
+import matplotlib.pyplot as plt
+from pandas_datareader import data as web
+from yahooquery import Ticker
+from statsmodels.tsa.api import VAR
+from statsmodels.tsa.vector_ar.vecm import VECM, coint_johansen
+from statsmodels.tsa.stattools import adfuller
+
+# ---------- CONFIGURAÇÃO DA PÁGINA ------------------------------------------
+st.set_page_config(
+    page_title="Choques Macroeconômicos – VAR • VECM • IRF",
+    layout="wide",
+    initial_sidebar_state="expanded"
+)
+
+# ---------- FUNÇÃO PARA BAIXAR DADOS ----------------------------------------
+@st.cache_data(show_spinner=False)
+def fetch_series(start='2000-01-01', end=None):
+    """
+    Baixa PIB real (trimestral), Core PCE (mensal) do FRED
+    e preço do petróleo WTI (diário) do YahooFinance;
+    devolve DataFrame log-nível, frequência mensal.
+    """
+    end = end or pd.Timestamp.today().strftime("%Y-%m-%d")
+
+    # --- FRED ---------------------------------------------------------------
+    fred = web.DataReader(['GDPC1', 'PCEPILFE'], 'fred', start, end)
+    pib_log  = np.log(fred['GDPC1']).to_period('M').to_timestamp('M')
+    pce_log  = np.log(fred['PCEPILFE']).resample('M').last()
+
+    # --- WTI ---------------------------------------------------------------
+    wti_raw = Ticker('CL=F').history(start=start, end=end)['adjclose'].droplevel(0)
+    wti_log = np.log(wti_raw).resample('M').last()
+
+    df = (
+        pd.concat([pib_log, pce_log, wti_log], axis=1)
+          .dropna()
+          .rename(columns={
+              'GDPC1'   : 'PIB_log',
+              'PCEPILFE': 'CORE_PCE_log',
+              'adjclose': 'WTI_log'
+          })
+    )
+    return df
+
+# ---------- SIDEBAR ---------------------------------------------------------
+st.sidebar.title("Controles do modelo")
+
+start_date = st.sidebar.date_input("Data inicial", pd.to_datetime("2000-01-01"))
+lags       = st.sidebar.slider("Número de lags (p)", 1, 12, 6)
+conf       = st.sidebar.slider("Nível de confiança (%)", 80, 99, 95)
+model_type = st.sidebar.selectbox("Modelo", ["VECM", "VAR"])
+show_fe    = st.sidebar.checkbox("Exibir FEVD", value=False)
+h_irf      = st.sidebar.slider("Horizonte IRF (meses)", 1, 36, 12)
+
+# ---------- CARREGAR DADOS --------------------------------------------------
+with st.spinner("Baixando séries…"):
+    df = fetch_series(start=start_date.strftime("%Y-%m-%d"))
+
+st.markdown("### Séries em log-nível (mensal)")
+st.line_chart(df)
+
+# ---------- TESTES ADF / JOHANSEN ------------------------------------------
+with st.expander("Testes de estacionariedade e cointegração", expanded=False):
+    col_test = st.columns(2)
+
+    with col_test[0]:
+        st.subheader("ADF (nível)")
+        for col in df.columns:
+            stat, p, *_ = adfuller(df[col])
+            st.write(f"**{col}**  ADF={stat:.2f}  p={p:.4f}")
+
+    with col_test[1]:
+        joh = coint_johansen(df, det_order=0, k_ar_diff=lags)
+        st.subheader("Johansen – Trace")
+        for i, (ts, cv) in enumerate(zip(joh.lr1, joh.cvt[:,1])):
+            decision = "Rejeita" if ts > cv else "Não rejeita"
+            st.write(f"r≤{i}: trace {ts:.2f} > crit5 {cv:.2f}? **{decision}**")
+
+# ---------- AJUSTE DO MODELO -----------------------------------------------
+if model_type == "VAR":
+    var = VAR(df).fit(lags)
+    st.success(f"VAR({lags}) ajustado.")
+else:
+    r = sum(joh.lr1 > joh.cvt[:,1])  # rank pelo trace
+    vecm = VECM(df, k_ar_diff=lags, coint_rank=r, deterministic='co').fit()
+    st.success(f"VECM ajustado: rank={r}, p={lags}")
+
+# ---------- IRF -------------------------------------------------------------
+with st.expander(" Impulse Response Function (IRF)", expanded=True):
+    impulse  = st.selectbox("Choque (impulse)", df.columns, index=2)
+    response = st.selectbox("Resposta", df.columns, index=0)
+    ic = 1 - conf/100
+
+    if model_type == "VAR":
+        irf = var.irf(h_irf).plot(impulse=impulse, response=response,
+                                  orth=False, signif=ic)
+    else:
+        irf = vecm.irf(h_irf).plot(impulse=impulse, response=response,
+                                   orth=False, signif=ic)
+
+    st.pyplot(irf.figure)
+
+# ---------- FEVD ------------------------------------------------------------
+if show_fe:
+    st.markdown("### Forecast-Error Variance Decomposition")
+    if model_type == "VAR":      # FEVD só p/ VAR em statsmodels 0.13
+        fevd = var.fevd(h_irf)
+    else:
+        fevd = vecm.fevd(h_irf)
+
+    horizons = [1, 6, 12, h_irf]
+    tbl = pd.concat(
+        {h: pd.DataFrame(fevd.decomp[h],
+                         columns=[f"Shock_{v}" for v in df.columns],
+                         index=[f"Var_{v}" for v in df.columns])
+         for h in horizons},
+        axis=0
+    ).round(3)*100
+    st.dataframe(tbl)
+
+# ---------- Rodapé ----------------------------------------------------------
+st.caption("Dados: FRED (GDP, Core PCE) & Yahoo Finance (WTI).  "
+           "Aplicação didática para Engenharia de Produtos Financeiros (UnB-FACE).")