diff options
| -rw-r--r-- | python/notebooks/dispersion_tranche_model.ipynb | 198 |
1 files changed, 27 insertions, 171 deletions
diff --git a/python/notebooks/dispersion_tranche_model.ipynb b/python/notebooks/dispersion_tranche_model.ipynb index 46eb348c..56255a42 100644 --- a/python/notebooks/dispersion_tranche_model.ipynb +++ b/python/notebooks/dispersion_tranche_model.ipynb @@ -18,7 +18,7 @@ "import serenitas.analytics.tranche_data as tdata\n", "\n", "from serenitas.analytics.basket_index import MarkitBasketIndex\n", - "from serenitas.analytics import on_the_run\n", + "from serenitas.analytics.index_data import on_the_run\n", "from statsmodels.graphics.regressionplots import plot_fit\n", "from scipy.special import logit, expit\n", "from serenitas.utils.db import dbengine, dbconn\n", @@ -52,117 +52,18 @@ { "cell_type": "code", "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [ - "#Section 1----------------------------------------------------\n", - "#index basis doesn't work with HY (opposite reaction to what I think)\n", - "#RFE\n", - "drop_variable_list = ['tranche_loss_per', 'tranche_id', 'index_price', 'detach', 'corr_at_detach', \n", - " 'corr01', 'exp_percentage', 'indexfactor', 'duration', 'index_expected_loss',\n", - " 'index_theta', 'delta', 'expected_loss', 'attach_adj', 'detach_adj',\n", - " 'cumulativeloss', \n", - " 'forward_delta', \n", - " #Comment out to include\n", - " # 'index_duration',\n", - " 'thickness',\n", - " 'moneyness',\n", - " # 'index_basis',\n", - " # 'att_moneyness', \n", - " # 'det_moneyness',\n", - " 'dispersion',\n", - " # 'gini', \n", - " 'gamma',\n", - " 'theta',\n", - " 'index_theta'\n", - " ]" - ] - }, - { - "cell_type": "code", - "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ - "def run_rfe(index_type):\n", - " risk = disp.get_tranche_data(dbconn(\"serenitasdb\"), index_type)\n", - " attach_max = risk.index.get_level_values(\"attach\").max()\n", - " bottom_stack = risk[risk.index.get_level_values(\"attach\") != attach_max]\n", - " bottom_stack = bottom_stack[bottom_stack.tranche_loss_per > 0].dropna()\n", - "\n", - " #prepare the variables\n", - " y = logit(bottom_stack['tranche_loss_per'])\n", - " X = bottom_stack.drop(drop_variable_list, axis=1)\n", - " \n", - " X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)\n", - " \n", - " pipe_rfe = make_pipeline (PowerTransformer(),\n", - " #PolynomialFeatures(degree=2),\n", - " PolynomialFeatures(interaction_only=True),\n", - " RFECV(estimator=LinearRegression(), \n", - " cv=10,\n", - " min_features_to_select=1))\n", - " \n", - " pipe_rfe.fit(X_train, y_train)\n", - " n_features_to_select = pipe_rfe['rfecv'].n_features_\n", - " pipe_rfe.steps[-1]= ('rfe', RFE(estimator=LinearRegression(), n_features_to_select = n_features_to_select))\n", - " model = pipe_rfe.fit(X_train, y_train)\n", - " \n", - " #RandomForest\n", - " #params = {'n_estimators': 100,\n", - " # 'min_samples_split': 3,\n", - " # 'verbose':1,\n", - " # 'n_jobs': -1}\n", - " #randomforest = RandomForestRegressor(**params)\n", - " \n", - " \n", - " #gradientboost\n", - " #params = {'n_estimators': 500,\n", - " # 'max_depth': 10,\n", - " # 'min_samples_split': 3,\n", - " # 'learning_rate': 0.01,\n", - " # 'loss': 'huber',\n", - " # 'verbose':1}\n", - " #gb = GradientBoostingRegressor(**params).fit(X_train, y_train)\n", - " \n", - " #model = VotingRegressor([('rf', model), ('gb', gb)]).fit(X_train, y_train)\n", - " #model = VotingRegressor([('lr', pipe_rfe)]).fit(X, logit(y))\n", - "\n", - " df = pd.merge(risk, \n", - " pd.DataFrame(expit(model.predict(X)), \n", - " index=X.index, \n", - " columns=['predict_tranche_loss']),\n", - " how='left', left_index=True, right_index=True)\n", - "\n", - " df.loc[df.index.get_level_values(\"attach\") != attach_max, \"predict_tranche_loss_per_index\"] = (\n", - " df.predict_tranche_loss * df.thickness / df.index_expected_loss\n", - " )\n", - "\n", - " def aux(s):\n", - " temp = s.values\n", - " temp[-1] = 1 - temp[:-1].sum()\n", - " return temp\n", - "\n", - " df[\"predict_tranche_loss_per_index\"] = df.groupby([\"index\", \"series\", \"date\"])[\"predict_tranche_loss_per_index\"].transform(aux)\n", - " df = df.assign(\n", - " mispricing=(df.exp_percentage - df.predict_tranche_loss_per_index)\n", - " * df.index_expected_loss\n", - " / (df.detach_adj - df.attach_adj)\n", - " )\n", - " rfe_result = pipe_rfe\n", - " print(index_type, \" num features: \", n_features_to_select)\n", - " print(index_type, \" Chosen columns: \", np.array(rfe_result['polynomialfeatures'].get_feature_names_out(X.columns))[rfe_result['rfe'].support_])\n", - " print(index_type, \" Training Score: \", model.score(X_train, y_train))\n", - " print(index_type, \" Testing Score: \", model.score(X_test, y_test))\n", - " \n", - " return model, df, X\n", - "\n", - "gini_model, gini_results, gini_X = {}, {}, {}\n", + "#Run RFE model\n", + "gini_model, gini_results = {}, {}\n", "fieldlist = ['exp_percentage','dispersion','gini','tranche_loss_per','mispricing']\n", "for index_type in ['HY', 'IG', 'EU', 'XO']:\n", - " gini_model[index_type], gini_results[index_type], gini_X[index_type] = run_rfe(index_type)\n", + " risk = disp.get_tranche_data(dbconn(\"serenitasdb\"), index_type)\n", + " risk = risk[risk.index_duration > 1] #filter out the short duration ones\n", + " gini_results[index_type], gini_model[index_type] = disp.create_rfe_models(risk)\n", " gini_results[index_type][fieldlist].to_csv('/home/serenitas/edwin/DispersionModel/' + index_type + '_results_rfecv.csv')" ] }, @@ -179,20 +80,27 @@ "for index_type in ['HY', 'IG', 'EU', 'XO']:\n", " plots = {}\n", " tranche_attach = []\n", - "\n", - " for i, X in gini_X[index_type].groupby('attach'):\n", + " \n", + " res = gini_results[index_type]\n", + " mod = gini_model[index_type]\n", + " \n", + " Xs = res[mod.feature_names_in_]\n", + " \n", + " for i, X in Xs.groupby('attach'):\n", " tranche_attach.append(X.index[0][5])\n", " for var in X.columns:\n", " bins = np.linspace(X[var].min(), X[var].max(),num=steps)\n", " testing_df = pd.DataFrame(bins, columns=[var])\n", " for var_1 in X.drop(var, axis=1).columns:\n", " testing_df = pd.concat([testing_df, pd.Series(np.repeat(X.iloc[-1][var_1], steps),name=var_1)], axis=1)\n", - " plots[i, var] = pd.Series(expit(gini_model[index_type].predict(testing_df[X.columns])), index=testing_df[var])\n", + " plots[i, var] = pd.Series(expit(mod.predict(testing_df[X.columns])), index=testing_df[var])\n", "\n", + " #breakpoint()\n", + " \n", " sensitivies = pd.concat(plots, names=['attach', 'shock', 'value'])\n", " sensitivies.to_csv('/home/serenitas/edwin/DispersionModel/' + index_type + '_sensitivies.csv')\n", "\n", - " fig, axes = plt.subplots(nrows=3, ncols=len(X.columns), figsize = (20,10))\n", + " fig, axes = plt.subplots(nrows=4, ncols=len(X.columns), figsize = (20,10))\n", " for i, p in enumerate(plots):\n", " x_loc = int(i/len(X.columns))\n", " y_loc = i % len(X.columns)\n", @@ -206,7 +114,7 @@ " rotation=90)\n", " fig.savefig(\"/home/serenitas/edwin/PythonGraphs/dispersion_model.png\", bbox_inches='tight')\n", "\n", - " fig_1, axes_1 = plt.subplots(nrows=3, ncols=1, figsize = (15,8))\n", + " fig_1, axes_1 = plt.subplots(nrows=4, ncols=1, figsize = (15,8))\n", " for i, p in enumerate(plots):\n", " x_loc = int(i/len(X.columns))\n", " plots[p].plot(ax=axes_1[x_loc], label=p[1], xlabel=\"\", legend=True)\n", @@ -234,7 +142,9 @@ { "cell_type": "code", "execution_count": null, - "metadata": {}, + "metadata": { + "tags": [] + }, "outputs": [], "source": [ "#Section 3----------------------------------------------------\n", @@ -259,7 +169,9 @@ { "cell_type": "code", "execution_count": null, - "metadata": {}, + "metadata": { + "tags": [] + }, "outputs": [], "source": [ "#plot the residuals\n", @@ -289,69 +201,13 @@ "data = risk[['gini', 'index_duration', 'index_expected_loss']]\n", "ols_model = smf.ols(\"gini ~ np.log(index_duration) + np.log(index_expected_loss)\", data=data).fit()\n" ] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [] } ], "metadata": { "kernelspec": { - "display_name": "Python 3.9.1 64-bit", + "display_name": "Python 3 (ipykernel)", "language": "python", - "name": "python39164bit6ddd573894c04d6a858a9a58880cc9d4" + "name": "python3" }, "language_info": { "codemirror_mode": { @@ -363,7 +219,7 @@ "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", - "version": "3.10.2" + "version": "3.10.8" } }, "nbformat": 4, |