def test_against_wls_inference(data, use_t, cov_type): y, x, w = data mod = RollingWLS(y, x, window=100, weights=w) res = mod.fit(use_t=use_t, cov_type=cov_type) ci = res.conf_int() # This is a smoke test of cov_params to make sure it works res.cov_params() # Skip to improve performance for i in range(100, y.shape[0]): _y = get_sub(y, i, 100) _x = get_sub(x, i, 100) wls = WLS(_y, _x, missing="drop").fit(use_t=use_t, cov_type=cov_type) assert_allclose(get_single(res.tvalues, i - 1), wls.tvalues) assert_allclose(get_single(res.bse, i - 1), wls.bse) assert_allclose(get_single(res.pvalues, i - 1), wls.pvalues, atol=1e-8) assert_allclose(get_single(res.fvalue, i - 1), wls.fvalue) with np.errstate(invalid="ignore"): assert_allclose(get_single(res.f_pvalue, i - 1), wls.f_pvalue, atol=1e-8) assert res.cov_type == wls.cov_type assert res.use_t == wls.use_t wls_ci = wls.conf_int() if isinstance(ci, pd.DataFrame): ci_val = ci.iloc[i - 1] ci_val = np.asarray(ci_val).reshape((-1, 2)) else: ci_val = ci[i - 1].T assert_allclose(ci_val, wls_ci)
def rolling_wls_model(): # Rolling Weighted Least Squares (Rolling WLS) from statsmodels.regression.rolling import RollingWLS data = get_dataset("longley") exog = sm.add_constant(data.exog, prepend=False) rolling_wls = RollingWLS(data.endog, exog) model = rolling_wls.fit(reset=50) return ModelWithResults(model=model, alg=rolling_wls, inference_dataframe=exog)
def test_error(): y, x, _ = gen_data(250, 2, True) with pytest.raises(ValueError, match="reset must be a positive integer"): RollingWLS(y, x,).fit(reset=-1) with pytest.raises(ValueError): RollingWLS(y, x).fit(method="unknown") with pytest.raises(ValueError, match="min_nobs must be larger"): RollingWLS(y, x, min_nobs=1) with pytest.raises(ValueError, match="min_nobs must be larger"): RollingWLS(y, x, window=60, min_nobs=100)
def test_formula(): y, x, w = gen_data(250, 3, True, pandas=True) fmla = "y ~ 1 + x0 + x1 + x2" data = pd.concat([y, x], axis=1) mod = RollingWLS.from_formula(fmla, window=100, data=data, weights=w) res = mod.fit() alt = RollingWLS(y, x, window=100) alt_res = alt.fit() assert_allclose(res.params, alt_res.params) ols_mod = RollingOLS.from_formula(fmla, window=100, data=data) ols_mod.fit()
def test_plot(): import matplotlib.pyplot as plt y, x, w = gen_data(250, 3, True, pandas=True) fmla = "y ~ 1 + x0 + x1 + x2" data = pd.concat([y, x], axis=1) mod = RollingWLS.from_formula(fmla, window=100, data=data, weights=w) res = mod.fit() fig = res.plot_recursive_coefficient() assert isinstance(fig, plt.Figure) res.plot_recursive_coefficient(variables=2, alpha=None, figsize=(30, 7)) res.plot_recursive_coefficient(variables="x0", alpha=None, figsize=(30, 7)) res.plot_recursive_coefficient(variables=[0, 2], alpha=None, figsize=(30, 7)) res.plot_recursive_coefficient(variables=["x0"], alpha=None, figsize=(30, 7)) res.plot_recursive_coefficient(variables=["x0", "x1", "x2"], alpha=None, figsize=(30, 7)) with pytest.raises(ValueError, match="variable x4 is not an integer"): res.plot_recursive_coefficient(variables="x4") fig = plt.Figure() # Just silence the warning with warnings.catch_warnings(): warnings.simplefilter("ignore") out = res.plot_recursive_coefficient(fig=fig) assert out is fig res.plot_recursive_coefficient(alpha=None, figsize=(30, 7))
def test_plot(): import matplotlib.pyplot as plt y, x, w = gen_data(250, 3, True, pandas=True) fmla = 'y ~ 1 + x0 + x1 + x2' data = pd.concat([y, x], axis=1) mod = RollingWLS.from_formula(fmla, window=100, data=data, weights=w) res = mod.fit() fig = res.plot_recursive_coefficient() assert isinstance(fig, plt.Figure) res.plot_recursive_coefficient(variables=2, alpha=None, figsize=(30, 7)) res.plot_recursive_coefficient(variables='x0', alpha=None, figsize=(30, 7)) res.plot_recursive_coefficient(variables=[0, 2], alpha=None, figsize=(30, 7)) res.plot_recursive_coefficient(variables=['x0'], alpha=None, figsize=(30, 7)) res.plot_recursive_coefficient(variables=['x0', 'x1', 'x2'], alpha=None, figsize=(30, 7)) with pytest.raises(ValueError, match='variable x4 is not an integer'): res.plot_recursive_coefficient(variables='x4') fig = plt.Figure() with pytest.warns(UserWarning, match="tight_layout"): out = res.plot_recursive_coefficient(fig=fig) assert out is fig res.plot_recursive_coefficient(alpha=None, figsize=(30, 7))
def test_has_nan(data): y, x, w = data mod = RollingWLS(y, x, window=100, weights=w) has_nan = np.zeros(y.shape[0], dtype=bool) for i in range(100, y.shape[0] + 1): _y = get_sub(y, i, 100) _x = get_sub(x, i, 100) has_nan[i - 1] = np.squeeze((np.any(np.isnan(_y)) or np.any(np.isnan(_x)))) assert_array_equal(mod._has_nan, has_nan)
def test_weighted_against_wls(weighted_data): y, x, w = weighted_data mod = RollingWLS(y, x, weights=w, window=100) res = mod.fit(use_t=True) for i in range(100, y.shape[0]): _y = get_sub(y, i, 100) _x = get_sub(x, i, 100) if w is not None: _w = get_sub(w, i, 100) else: _w = np.ones_like(_y) wls = WLS(_y, _x, weights=_w, missing="drop").fit() rolling_params = get_single(res.params, i - 1) rolling_nobs = get_single(res.nobs, i - 1) assert_allclose(rolling_params, wls.params) assert_allclose(rolling_nobs, wls.nobs) assert_allclose(get_single(res.ssr, i - 1), wls.ssr) assert_allclose(get_single(res.llf, i - 1), wls.llf) assert_allclose(get_single(res.aic, i - 1), wls.aic) assert_allclose(get_single(res.bic, i - 1), wls.bic) assert_allclose(get_single(res.centered_tss, i - 1), wls.centered_tss) assert_allclose(res.df_model, wls.df_model) assert_allclose(get_single(res.df_resid, i - 1), wls.df_resid) assert_allclose(get_single(res.ess, i - 1), wls.ess, atol=1e-8) assert_allclose(res.k_constant, wls.k_constant) assert_allclose(get_single(res.mse_model, i - 1), wls.mse_model) assert_allclose(get_single(res.mse_resid, i - 1), wls.mse_resid) assert_allclose(get_single(res.mse_total, i - 1), wls.mse_total) assert_allclose( get_single(res.rsquared, i - 1), wls.rsquared, atol=1e-8 ) assert_allclose( get_single(res.rsquared_adj, i - 1), wls.rsquared_adj, atol=1e-8 ) assert_allclose( get_single(res.uncentered_tss, i - 1), wls.uncentered_tss )
def test_raise(data): y, x, w = data mod = RollingWLS(y, x, window=100, missing="drop", weights=w) res = mod.fit() params = np.asarray(res.params) assert np.all(np.isfinite(params[99:])) if not np.any(np.isnan(y)): return mod = RollingWLS(y, x, window=100, missing="skip") res = mod.fit() params = np.asarray(res.params) assert np.any(np.isnan(params[100:]))