{ "cells": [ { "cell_type": "code", "execution_count": 1, "id": "b1eb1116-fe0d-4c53-ab06-6d52bd64b716", "metadata": {}, "outputs": [], "source": [ "import pandas as pd\n", "import pandas as pd\n", "from sklearn.preprocessing import StandardScaler\n", "from sklearn.ensemble import RandomForestRegressor\n", "from sklearn.model_selection import train_test_split" ] }, { "cell_type": "code", "execution_count": 2, "id": "76221885-2345-4447-a360-26e5fda94c43", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ " CO_ID Name Year Quarter Sector Return on Assets Return on Capital \\\n", "0 2030.0 SARCO 2015 1 Energy 1.236 0.265 \n", "1 2030.0 SARCO 2015 2 Energy 1.246 0.703 \n", "2 2030.0 SARCO 2015 3 Energy 1.143 0.792 \n", "3 2030.0 SARCO 2015 4 Energy NaN NaN \n", "4 2030.0 SARCO 2016 1 Energy 1.520 0.860 \n", "\n", " Return on Equity Return on Common Equity Gross Profit Margin ... \\\n", "0 1.960 1.24 100.0 ... \n", "1 1.950 1.65 100.0 ... \n", "2 1.960 -1.12 100.0 ... \n", "3 1.930 -1.23 100.0 ... \n", "4 1.240 2.65 100.0 ... \n", "\n", " Asset Turnover Current Ratio Debt/Equity Long Term Debt/Captial \\\n", "0 0.03 3.39 0.066 0.033 \n", "1 0.02 3.38 NaN NaN \n", "2 0.06 3.35 NaN 0.021 \n", "3 0.04 3.39 0.026 NaN \n", "4 -1.00 2.90 NaN NaN \n", "\n", " Book Value Per Share Start Price End Price Stock Return TASI Return \\\n", "0 19.190 50.5 57.00 0.128713 0.053471 \n", "1 19.560 57.0 58.75 0.030702 0.080545 \n", "2 19.730 58.5 41.00 -0.299145 -0.081079 \n", "3 19.760 41.0 41.10 0.002439 -0.263944 \n", "4 23.650 41.0 33.50 -0.182927 0.037782 \n", "\n", " Relative_Return \n", "0 0.075242 \n", "1 -0.049844 \n", "2 -0.218066 \n", "3 0.266383 \n", "4 -0.220708 \n", "\n", "[5 rows x 23 columns]\n" ] } ], "source": [ "# Load the dataset\n", "df = pd.read_csv('datasetNewNewNewTASI.csv')\n", "print(df.head())" ] }, { "cell_type": "code", "execution_count": 3, "id": "91e7b57b-bb0c-498d-84e5-a26742197bf4", "metadata": {}, "outputs": [ { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
CO_IDNameYearQuarterSectorReturn on AssetsReturn on CapitalReturn on EquityReturn on Common EquityGross Profit Margin...Asset TurnoverCurrent RatioDebt/EquityLong Term Debt/CaptialBook Value Per ShareStart PriceEnd PriceStock ReturnTASI ReturnRelative_Return
02030.0SARCO20151Energy1.2360.2651.9601.24100.0...0.033.390.0660.03319.19050.557.000.1287130.0534710.075242
12030.0SARCO20152Energy1.2460.7031.9501.65100.0...0.023.38NaNNaN19.56057.058.750.0307020.080545-0.049844
22030.0SARCO20153Energy1.1430.7921.960-1.12100.0...0.063.35NaN0.02119.73058.541.00-0.299145-0.081079-0.218066
32030.0SARCO20154EnergyNaNNaN1.930-1.23100.0...0.043.390.026NaN19.76041.041.100.002439-0.2639440.266383
42030.0SARCO20161Energy1.5200.8601.2402.65100.0...-1.002.90NaNNaN23.65041.033.50-0.1829270.037782-0.220708
\n", "

5 rows × 23 columns

\n", "
" ], "text/plain": [ " CO_ID Name Year Quarter Sector Return on Assets Return on Capital \\\n", "0 2030.0 SARCO 2015 1 Energy 1.236 0.265 \n", "1 2030.0 SARCO 2015 2 Energy 1.246 0.703 \n", "2 2030.0 SARCO 2015 3 Energy 1.143 0.792 \n", "3 2030.0 SARCO 2015 4 Energy NaN NaN \n", "4 2030.0 SARCO 2016 1 Energy 1.520 0.860 \n", "\n", " Return on Equity Return on Common Equity Gross Profit Margin ... \\\n", "0 1.960 1.24 100.0 ... \n", "1 1.950 1.65 100.0 ... \n", "2 1.960 -1.12 100.0 ... \n", "3 1.930 -1.23 100.0 ... \n", "4 1.240 2.65 100.0 ... \n", "\n", " Asset Turnover Current Ratio Debt/Equity Long Term Debt/Captial \\\n", "0 0.03 3.39 0.066 0.033 \n", "1 0.02 3.38 NaN NaN \n", "2 0.06 3.35 NaN 0.021 \n", "3 0.04 3.39 0.026 NaN \n", "4 -1.00 2.90 NaN NaN \n", "\n", " Book Value Per Share Start Price End Price Stock Return TASI Return \\\n", "0 19.190 50.5 57.00 0.128713 0.053471 \n", "1 19.560 57.0 58.75 0.030702 0.080545 \n", "2 19.730 58.5 41.00 -0.299145 -0.081079 \n", "3 19.760 41.0 41.10 0.002439 -0.263944 \n", "4 23.650 41.0 33.50 -0.182927 0.037782 \n", "\n", " Relative_Return \n", "0 0.075242 \n", "1 -0.049844 \n", "2 -0.218066 \n", "3 0.266383 \n", "4 -0.220708 \n", "\n", "[5 rows x 23 columns]" ] }, "execution_count": 3, "metadata": {}, "output_type": "execute_result" } ], "source": [ "df.head()" ] }, { "cell_type": "code", "execution_count": 4, "id": "9212ecfe-b3d9-433b-8b11-909cd0634952", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Long Term Debt/Captial 299\n", "Debt/Equity 264\n", "Net Income Margin 184\n", "EBITA Margin 182\n", "Gross Profit Margin 174\n", "EBIT Margin 173\n", "Return on Common Equity 162\n", "Asset Turnover 160\n", "Return on Equity 159\n", "Return on Assets 142\n", "Return on Capital 140\n", "Current Ratio 140\n", "Book Value Per Share 62\n", "dtype: int64\n" ] } ], "source": [ "missing = df.isnull().sum()\n", "missing = missing[missing > 0].sort_values(ascending=False)\n", "print(missing)" ] }, { "cell_type": "code", "execution_count": 5, "id": "b3f14bf0-a379-408d-bd8e-93c6fe04b2c3", "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "C:\\Users\\يوسف عثمان\\AppData\\Local\\Temp\\ipykernel_32524\\3887843883.py:10: UserWarning: Could not infer format, so each element will be parsed individually, falling back to `dateutil`. To ensure parsing is consistent and as-expected, please specify a format.\n", " df[\"Quarter\"] = pd.to_datetime(df[\"Year\"].astype(str) + \"Q\" + df[\"Quarter\"].astype(str))\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "✅ Missing values handled and Z-score standardization applied.\n", "📌 Features: ['Return on Assets', 'Return on Capital', 'Return on Equity', 'Return on Common Equity', 'Gross Profit Margin', 'EBITA Margin', 'EBIT Margin', 'Net Income Margin', 'Asset Turnover', 'Current Ratio', 'Debt/Equity', 'Long Term Debt/Captial', 'Book Value Per Share']\n", "🎯 Target: Relative_Return\n" ] } ], "source": [ "import pandas as pd\n", "import numpy as np\n", "\n", "# Load your dataset\n", "df = pd.read_csv(\"datasetNewNewNewTASI.csv\")\n", "\n", "# Step 1: Strip column names of whitespace\n", "df.columns = df.columns.str.strip()\n", "\n", "df[\"Quarter\"] = pd.to_datetime(df[\"Year\"].astype(str) + \"Q\" + df[\"Quarter\"].astype(str))\n", "\n", "\n", "# Step 2: Define target and features\n", "target_col = 'Relative_Return'\n", "feature_cols = [\n", " 'Return on Assets', 'Return on Capital', 'Return on Equity', 'Return on Common Equity',\n", " 'Gross Profit Margin', 'EBITA Margin', 'EBIT Margin', 'Net Income Margin',\n", " 'Asset Turnover', 'Current Ratio', 'Debt/Equity', 'Long Term Debt/Captial',\n", " 'Book Value Per Share'\n", "]\n", "\n", "# Step 3: Fill missing values with forward fill then backward fill by company\n", "# Exclude grouping column (CO_ID) from being operated on in .apply()\n", "cols_to_fill = feature_cols + [target_col]\n", "df[cols_to_fill] = (\n", " df.groupby(\"CO_ID\")[cols_to_fill]\n", " .apply(lambda group: group.ffill().bfill())\n", " .reset_index(drop=True)\n", ")\n", "\n", "# Step 4: Convert feature columns to numeric (in case of object type)\n", "df[feature_cols] = df[feature_cols].apply(pd.to_numeric, errors='coerce')\n", "\n", "df[feature_cols] = df[feature_cols].fillna(df[feature_cols].mean())\n", "\n", "# Step 5: Apply Z-score normalization from scratch\n", "for col in feature_cols:\n", " mean = df[col].mean()\n", " std = df[col].std()\n", " df[col] = (df[col] - mean) / std\n", "\n", "# ✅ Done\n", "print(\"✅ Missing values handled and Z-score standardization applied.\")\n", "print(\"📌 Features:\", feature_cols)\n", "print(\"🎯 Target:\", target_col)\n" ] }, { "cell_type": "code", "execution_count": 6, "id": "d162759c-64fb-4ca3-8c3c-79b6ccce01cb", "metadata": {}, "outputs": [ { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
CO_IDNameYearQuarterSectorReturn on AssetsReturn on CapitalReturn on EquityReturn on Common EquityGross Profit Margin...Asset TurnoverCurrent RatioDebt/EquityLong Term Debt/CaptialBook Value Per ShareStart PriceEnd PriceStock ReturnTASI ReturnRelative_Return
02030.0SARCO20152015-01-01Energy-0.252465-0.239364-0.358703-0.316179-0.474507...-0.001501-0.4304211.0955491.2493240.23497650.557.000.1287130.0534710.247680
12030.0SARCO20152015-04-01Energy-0.258880-0.482341-0.248664-0.219191-0.474088...-0.001501-0.4888531.0582661.2493240.22878257.058.750.0307020.0805450.086175
22030.0SARCO20152015-07-01Energy-0.249258-0.315901-0.393915-0.347215-0.462763...-0.027798-0.4888530.9329411.3447770.19244358.541.00-0.299145-0.081079-0.324735
32030.0SARCO20152015-10-01Energy-0.250862-0.245438-0.364865-0.321610-0.464861...-0.027798-0.4532861.1018421.2058260.25975341.041.100.002439-0.2639440.106975
42030.0SARCO20162016-01-01Energy-0.242843-0.253942-0.467861-0.412391-0.431306...-0.029990-0.4532860.4125911.2058260.07310341.033.50-0.1829270.037782-0.424714
\n", "

5 rows × 23 columns

\n", "
" ], "text/plain": [ " CO_ID Name Year Quarter Sector Return on Assets \\\n", "0 2030.0 SARCO 2015 2015-01-01 Energy -0.252465 \n", "1 2030.0 SARCO 2015 2015-04-01 Energy -0.258880 \n", "2 2030.0 SARCO 2015 2015-07-01 Energy -0.249258 \n", "3 2030.0 SARCO 2015 2015-10-01 Energy -0.250862 \n", "4 2030.0 SARCO 2016 2016-01-01 Energy -0.242843 \n", "\n", " Return on Capital Return on Equity Return on Common Equity \\\n", "0 -0.239364 -0.358703 -0.316179 \n", "1 -0.482341 -0.248664 -0.219191 \n", "2 -0.315901 -0.393915 -0.347215 \n", "3 -0.245438 -0.364865 -0.321610 \n", "4 -0.253942 -0.467861 -0.412391 \n", "\n", " Gross Profit Margin ... Asset Turnover Current Ratio Debt/Equity \\\n", "0 -0.474507 ... -0.001501 -0.430421 1.095549 \n", "1 -0.474088 ... -0.001501 -0.488853 1.058266 \n", "2 -0.462763 ... -0.027798 -0.488853 0.932941 \n", "3 -0.464861 ... -0.027798 -0.453286 1.101842 \n", "4 -0.431306 ... -0.029990 -0.453286 0.412591 \n", "\n", " Long Term Debt/Captial Book Value Per Share Start Price End Price \\\n", "0 1.249324 0.234976 50.5 57.00 \n", "1 1.249324 0.228782 57.0 58.75 \n", "2 1.344777 0.192443 58.5 41.00 \n", "3 1.205826 0.259753 41.0 41.10 \n", "4 1.205826 0.073103 41.0 33.50 \n", "\n", " Stock Return TASI Return Relative_Return \n", "0 0.128713 0.053471 0.247680 \n", "1 0.030702 0.080545 0.086175 \n", "2 -0.299145 -0.081079 -0.324735 \n", "3 0.002439 -0.263944 0.106975 \n", "4 -0.182927 0.037782 -0.424714 \n", "\n", "[5 rows x 23 columns]" ] }, "execution_count": 6, "metadata": {}, "output_type": "execute_result" } ], "source": [ "df.head()" ] }, { "cell_type": "code", "execution_count": 7, "id": "c6da65e1-a740-4411-9947-0676f9dee88e", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Series([], dtype: int64)\n" ] } ], "source": [ "missing = df.isnull().sum()\n", "missing = missing[missing > 0].sort_values(ascending=False)\n", "print(missing)" ] }, { "cell_type": "code", "execution_count": 863, "id": "0fbcdcf5-b5be-41d7-9278-a1a3e4f8dec9", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "✅ Rolling window sequences created.\n", "📏 X shape (samples, timesteps, features): (1850, 4, 13)\n", "🎯 y shape (samples,): (1850,)\n" ] } ], "source": [ "import numpy as np\n", "\n", "# Set rolling window size\n", "window_size = 4\n", "\n", "# Create empty lists to hold sequences and labels\n", "X_sequences = []\n", "y_labels = []\n", "\n", "# Sort the dataframe by company and quarter to maintain time order\n", "df = df.sort_values(by=[\"CO_ID\", \"Quarter\"])\n", "\n", "# Loop over each company separately\n", "for co_id, group in df.groupby(\"CO_ID\"):\n", " group = group.reset_index(drop=True)\n", " \n", " # Skip if not enough quarters\n", " if len(group) <= window_size:\n", " continue\n", "\n", " for i in range(len(group) - window_size):\n", " # Create sequence of features\n", " X_seq = group.loc[i:i+window_size-1, feature_cols].values.astype(np.float32)\n", " # Target is the value in the next quarter\n", " y_label = group.loc[i + window_size, target_col]\n", "\n", " X_sequences.append(X_seq)\n", " y_labels.append(y_label)\n", "\n", "# Convert to numpy arrays\n", "X_sequences = np.array(X_sequences)\n", "y_labels = np.array(y_labels).astype(np.float32)\n", "\n", "# Confirm shapes\n", "print(\"✅ Rolling window sequences created.\")\n", "print(\"📏 X shape (samples, timesteps, features):\", X_sequences.shape)\n", "print(\"🎯 y shape (samples,):\", y_labels.shape)\n", "\n" ] }, { "cell_type": "code", "execution_count": 864, "id": "b25e9b1a-5488-48b2-a06d-1df0d5863331", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Company: 1201.0\n", "Dates in sequence: [Timestamp('2015-01-01 00:00:00'), Timestamp('2015-04-01 00:00:00'), Timestamp('2015-07-01 00:00:00'), Timestamp('2015-10-01 00:00:00')]\n", "Target date: 2016-01-01 00:00:00\n" ] } ], "source": [ "df[\"Quarter\"] = pd.to_datetime(df[\"Quarter\"], dayfirst=True)\n", "df = df.sort_values(by=[\"CO_ID\", \"Quarter\"])\n", "\n", "# Example: Print dates used for first sequence in first company\n", "for co_id, group in df.groupby(\"CO_ID\"):\n", " group = group.reset_index(drop=True)\n", " print(\"Company:\", co_id)\n", " print(\"Dates in sequence:\", group.loc[0:3, \"Quarter\"].tolist())\n", " print(\"Target date:\", group.loc[4, \"Quarter\"])\n", " break # Remove this if you want all" ] }, { "cell_type": "code", "execution_count": 865, "id": "d9042888-0db6-4808-94b5-7f15ef9c400c", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\n", "['2015-01-01 00:00:00', '2015-04-01 00:00:00', '2015-07-01 00:00:00',\n", " '2015-10-01 00:00:00', '2016-01-01 00:00:00', '2016-04-01 00:00:00',\n", " '2016-07-01 00:00:00', '2016-10-01 00:00:00', '2017-01-01 00:00:00',\n", " '2017-04-01 00:00:00']\n", "Length: 10, dtype: datetime64[ns]\n" ] } ], "source": [ "print(df[\"Quarter\"].unique()[:10])\n" ] }, { "cell_type": "code", "execution_count": 866, "id": "c1ac56fa-8973-4ffc-b2a9-7d6185582cae", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "123 2015-01-01\n", "124 2015-04-01\n", "125 2015-07-01\n", "126 2015-10-01\n", "127 2016-01-01\n", "Name: Quarter, dtype: datetime64[ns]\n" ] } ], "source": [ "print(df[\"Quarter\"].head())\n" ] }, { "cell_type": "code", "execution_count": 867, "id": "924836da-e17a-412f-af49-0651f603087c", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "✅ Train shapes: (1480, 4, 13) (1480,)\n", "✅ Test shapes: (370, 4, 13) (370,)\n" ] } ], "source": [ "from sklearn.model_selection import train_test_split\n", "\n", "# Convert everything to numpy (if not already done)\n", "X = np.array(X_sequences)\n", "y = np.array(y_labels)\n", "\n", "# Determine split index (e.g., 80% train, 20% test)\n", "split_idx = int(len(X) * 0.8)\n", "\n", "# Split\n", "X_train, X_test = X[:split_idx], X[split_idx:]\n", "y_train, y_test = y[:split_idx], y[split_idx:]\n", "\n", "# Print shapes\n", "print(\"✅ Train shapes:\", X_train.shape, y_train.shape)\n", "print(\"✅ Test shapes:\", X_test.shape, y_test.shape)\n" ] }, { "cell_type": "code", "execution_count": 868, "id": "65d17d4c-6786-4629-a0c4-8b15b83fbfb3", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "✅ DataLoaders created.\n", "Train: 1295 samples\n", "Validation: 277 samples\n", "Test: 278 samples\n" ] } ], "source": [ "import torch\n", "from torch.utils.data import Dataset, DataLoader, random_split\n", "\n", "# Define custom PyTorch Dataset\n", "class StockDataset(Dataset):\n", " def __init__(self, X, y):\n", " self.X = torch.tensor(X, dtype=torch.float32)\n", " self.y = torch.tensor(y, dtype=torch.float32)\n", "\n", " def __len__(self):\n", " return len(self.X)\n", "\n", " def __getitem__(self, idx):\n", " return self.X[idx], self.y[idx]\n", "\n", "# Create dataset\n", "dataset = StockDataset(X_sequences, y_labels)\n", "\n", "# Split sizes\n", "total_size = len(dataset)\n", "train_size = int(0.7 * total_size)\n", "val_size = int(0.15 * total_size)\n", "test_size = total_size - train_size - val_size\n", "\n", "# Split dataset with fixed random seed for reproducibility\n", "train_dataset, val_dataset, test_dataset = random_split(\n", " dataset, [train_size, val_size, test_size],\n", " generator=torch.Generator().manual_seed(42)\n", ")\n", "\n", "# Create DataLoaders\n", "batch_size = 128\n", "train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)\n", "val_loader = DataLoader(val_dataset, batch_size=batch_size, shuffle=False)\n", "test_loader = DataLoader(test_dataset, batch_size=batch_size, shuffle=False)\n", "\n", "# Summary\n", "print(\"✅ DataLoaders created.\")\n", "print(f\"Train: {len(train_dataset)} samples\")\n", "print(f\"Validation: {len(val_dataset)} samples\")\n", "print(f\"Test: {len(test_dataset)} samples\")\n" ] }, { "cell_type": "code", "execution_count": 869, "id": "f84c3dfb-01c0-4d6b-a4c1-ec337b110fac", "metadata": {}, "outputs": [], "source": [ "# Hyperparameters (tune here)\n", "params = {\n", " \"input_size\": 13, # Number of features\n", " \"hidden_size\": 128,\n", " \"num_layers\": 2,\n", " \"dropout\": 0.4,\n", " \"batch_size\": 128,\n", " \"num_epochs\": 100,\n", " \"learning_rate\": 0.0005\n", "}\n" ] }, { "cell_type": "code", "execution_count": 870, "id": "bb17e81e-7bfd-409c-ba72-75c13994e913", "metadata": {}, "outputs": [], "source": [ "import torch\n", "from torch.utils.data import TensorDataset, DataLoader\n", "\n", "# Convert to tensors\n", "X_tensor = torch.tensor(X_sequences, dtype=torch.float32)\n", "y_tensor = torch.tensor(y_labels, dtype=torch.float32)\n", "\n", "#y_tensor = torch.clamp(y_tensor, min=-2.0, max=2.0)\n", "\n", "# Split 80% train, 20% val\n", "train_size = int(0.8 * len(X_tensor))\n", "X_train, X_val = X_tensor[:train_size], X_tensor[train_size:]\n", "y_train, y_val = y_tensor[:train_size], y_tensor[train_size:]\n", "\n", "# Datasets & Loaders\n", "train_dataset = TensorDataset(X_train, y_train)\n", "val_dataset = TensorDataset(X_val, y_val)\n", "\n", "train_loader = DataLoader(train_dataset, batch_size=params[\"batch_size\"], shuffle=True)\n", "val_loader = DataLoader(val_dataset, batch_size=params[\"batch_size\"], shuffle=False)\n" ] }, { "cell_type": "code", "execution_count": 871, "id": "01383e06-fc84-455f-a77e-90354460c619", "metadata": {}, "outputs": [], "source": [ "import torch.nn as nn\n", "\n", "class LSTMModel(nn.Module):\n", " def __init__(self, input_size, hidden_size, num_layers, dropout):\n", " super(LSTMModel, self).__init__()\n", " self.lstm = nn.LSTM(input_size, hidden_size, num_layers,\n", " dropout=dropout, batch_first=True)\n", " self.fc = nn.Linear(hidden_size, 1)\n", "\n", " #self.fc = nn.Sequential(\n", " #nn.Linear(hidden_size, 1),\n", " #nn.Sigmoid()\n", " #)\n", "\n", "\n", " def forward(self, x):\n", " lstm_out, _ = self.lstm(x)\n", " out = lstm_out[:, -1, :] # last time step\n", " return self.fc(out)\n" ] }, { "cell_type": "code", "execution_count": 872, "id": "ca665f76-5e04-45a5-a329-38d3bda0f4ff", "metadata": {}, "outputs": [], "source": [ "from torch.optim.lr_scheduler import ReduceLROnPlateau\n", "\n", "# Instantiate model\n", "model = LSTMModel(params[\"input_size\"], params[\"hidden_size\"],\n", " params[\"num_layers\"], params[\"dropout\"])\n", "criterion = nn.MSELoss()\n", "#criterion = nn.SmoothL1Loss() # Huber loss\n", "\n", "optimizer = torch.optim.Adam(model.parameters(), lr=params[\"learning_rate\"])\n", "#optimizer = torch.optim.Adam(model.parameters(), lr=params[\"learning_rate\"], weight_decay=0.0005)\n", "\n", "\n", "#scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='min', factor=0.5, patience=10, min_lr=0.0005)\n", "\n", "#scheduler = ReduceLROnPlateau(optimizer, mode='min', factor=0.1, patience=10)\n", "\n", "\n", "\n" ] }, { "cell_type": "code", "execution_count": 873, "id": "9371f3f0-d9ee-498b-a995-ee623ce600fb", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Epoch 1/100 | Train Loss: 0.1238 | Val Loss: 0.0721\n", "Epoch 2/100 | Train Loss: 0.1232 | Val Loss: 0.0727\n", "Epoch 3/100 | Train Loss: 0.1827 | Val Loss: 0.0727\n", "Epoch 4/100 | Train Loss: 0.1234 | Val Loss: 0.0730\n", "Epoch 5/100 | Train Loss: 0.1214 | Val Loss: 0.0734\n", "Epoch 6/100 | Train Loss: 0.1220 | Val Loss: 0.0738\n", "Epoch 7/100 | Train Loss: 0.1220 | Val Loss: 0.0741\n", "Epoch 8/100 | Train Loss: 0.1213 | Val Loss: 0.0741\n", "Epoch 9/100 | Train Loss: 0.1792 | Val Loss: 0.0742\n", "Epoch 10/100 | Train Loss: 0.1786 | Val Loss: 0.0748\n", "Epoch 11/100 | Train Loss: 0.1200 | Val Loss: 0.0758\n", "Epoch 12/100 | Train Loss: 0.1197 | Val Loss: 0.0778\n", "Epoch 13/100 | Train Loss: 0.1198 | Val Loss: 0.0768\n", "Epoch 14/100 | Train Loss: 0.1205 | Val Loss: 0.0760\n", "Epoch 15/100 | Train Loss: 0.1188 | Val Loss: 0.0763\n", "Epoch 16/100 | Train Loss: 0.1179 | Val Loss: 0.0764\n", "Epoch 17/100 | Train Loss: 0.1181 | Val Loss: 0.0757\n", "Epoch 18/100 | Train Loss: 0.1181 | Val Loss: 0.0760\n", "Epoch 19/100 | Train Loss: 0.1173 | Val Loss: 0.0753\n", "Epoch 20/100 | Train Loss: 0.1157 | Val Loss: 0.0753\n", "Epoch 21/100 | Train Loss: 0.1159 | Val Loss: 0.0750\n", "Epoch 22/100 | Train Loss: 0.1149 | Val Loss: 0.0765\n", "Epoch 23/100 | Train Loss: 0.1136 | Val Loss: 0.0756\n", "Epoch 24/100 | Train Loss: 0.1141 | Val Loss: 0.0736\n", "Epoch 25/100 | Train Loss: 0.1689 | Val Loss: 0.0745\n", "Epoch 26/100 | Train Loss: 0.1113 | Val Loss: 0.0748\n", "Epoch 27/100 | Train Loss: 0.1132 | Val Loss: 0.0765\n", "Epoch 28/100 | Train Loss: 0.1133 | Val Loss: 0.0741\n", "Epoch 29/100 | Train Loss: 0.1105 | Val Loss: 0.0731\n", "Epoch 30/100 | Train Loss: 0.1106 | Val Loss: 0.0734\n", "Epoch 31/100 | Train Loss: 0.1051 | Val Loss: 0.0748\n", "Epoch 32/100 | Train Loss: 0.1096 | Val Loss: 0.0755\n", "Epoch 33/100 | Train Loss: 0.1139 | Val Loss: 0.0727\n", "Epoch 34/100 | Train Loss: 0.1561 | Val Loss: 0.0723\n", "Epoch 35/100 | Train Loss: 0.1079 | Val Loss: 0.0748\n", "Epoch 36/100 | Train Loss: 0.0993 | Val Loss: 0.0748\n", "Epoch 37/100 | Train Loss: 0.1082 | Val Loss: 0.0726\n", "Epoch 38/100 | Train Loss: 0.1018 | Val Loss: 0.0727\n", "Epoch 39/100 | Train Loss: 0.1101 | Val Loss: 0.0732\n", "Epoch 40/100 | Train Loss: 0.0950 | Val Loss: 0.0739\n", "Epoch 41/100 | Train Loss: 0.0936 | Val Loss: 0.0735\n", "Epoch 42/100 | Train Loss: 0.1075 | Val Loss: 0.0740\n", "Epoch 43/100 | Train Loss: 0.1009 | Val Loss: 0.0729\n", "Epoch 44/100 | Train Loss: 0.0907 | Val Loss: 0.0731\n", "Epoch 45/100 | Train Loss: 0.0921 | Val Loss: 0.0742\n", "Epoch 46/100 | Train Loss: 0.0978 | Val Loss: 0.0744\n", "Epoch 47/100 | Train Loss: 0.0833 | Val Loss: 0.0732\n", "Epoch 48/100 | Train Loss: 0.0934 | Val Loss: 0.0735\n", "Epoch 49/100 | Train Loss: 0.0814 | Val Loss: 0.0735\n", "Epoch 50/100 | Train Loss: 0.0900 | Val Loss: 0.0742\n", "Epoch 51/100 | Train Loss: 0.0893 | Val Loss: 0.0743\n", "Epoch 52/100 | Train Loss: 0.0829 | Val Loss: 0.0738\n", "Epoch 53/100 | Train Loss: 0.0809 | Val Loss: 0.0738\n", "Epoch 54/100 | Train Loss: 0.0775 | Val Loss: 0.0732\n", "Epoch 55/100 | Train Loss: 0.0796 | Val Loss: 0.0739\n", "Epoch 56/100 | Train Loss: 0.0883 | Val Loss: 0.0744\n", "Epoch 57/100 | Train Loss: 0.0674 | Val Loss: 0.0731\n", "Epoch 58/100 | Train Loss: 0.0753 | Val Loss: 0.0731\n", "Epoch 59/100 | Train Loss: 0.0735 | Val Loss: 0.0728\n", "Epoch 60/100 | Train Loss: 0.0820 | Val Loss: 0.0729\n", "Epoch 61/100 | Train Loss: 0.0670 | Val Loss: 0.0739\n", "Epoch 62/100 | Train Loss: 0.0601 | Val Loss: 0.0724\n", "Epoch 63/100 | Train Loss: 0.0578 | Val Loss: 0.0720\n", "Epoch 64/100 | Train Loss: 0.0607 | Val Loss: 0.0724\n", "Epoch 65/100 | Train Loss: 0.0565 | Val Loss: 0.0729\n", "Epoch 66/100 | Train Loss: 0.0535 | Val Loss: 0.0722\n", "Epoch 67/100 | Train Loss: 0.0513 | Val Loss: 0.0718\n", "Epoch 68/100 | Train Loss: 0.0519 | Val Loss: 0.0721\n", "Epoch 69/100 | Train Loss: 0.0475 | Val Loss: 0.0722\n", "Epoch 70/100 | Train Loss: 0.0464 | Val Loss: 0.0716\n", "Epoch 71/100 | Train Loss: 0.0471 | Val Loss: 0.0718\n", "Epoch 72/100 | Train Loss: 0.0450 | Val Loss: 0.0719\n", "Epoch 73/100 | Train Loss: 0.0442 | Val Loss: 0.0717\n", "Epoch 74/100 | Train Loss: 0.0433 | Val Loss: 0.0718\n", "Epoch 75/100 | Train Loss: 0.0422 | Val Loss: 0.0718\n", "Epoch 76/100 | Train Loss: 0.0442 | Val Loss: 0.0720\n", "Epoch 77/100 | Train Loss: 0.0433 | Val Loss: 0.0720\n", "Epoch 78/100 | Train Loss: 0.0412 | Val Loss: 0.0717\n", "Epoch 79/100 | Train Loss: 0.0422 | Val Loss: 0.0719\n", "Epoch 80/100 | Train Loss: 0.0427 | Val Loss: 0.0719\n", "Epoch 81/100 | Train Loss: 0.0420 | Val Loss: 0.0719\n", "Epoch 82/100 | Train Loss: 0.0415 | Val Loss: 0.0721\n", "Epoch 83/100 | Train Loss: 0.0431 | Val Loss: 0.0719\n", "Epoch 84/100 | Train Loss: 0.0421 | Val Loss: 0.0723\n", "Epoch 85/100 | Train Loss: 0.0406 | Val Loss: 0.0726\n", "Epoch 86/100 | Train Loss: 0.0412 | Val Loss: 0.0724\n", "Epoch 87/100 | Train Loss: 0.0409 | Val Loss: 0.0723\n", "Epoch 88/100 | Train Loss: 0.0431 | Val Loss: 0.0721\n", "Epoch 89/100 | Train Loss: 0.0412 | Val Loss: 0.0719\n", "Epoch 90/100 | Train Loss: 0.0406 | Val Loss: 0.0722\n", "Epoch 91/100 | Train Loss: 0.0407 | Val Loss: 0.0723\n", "Epoch 92/100 | Train Loss: 0.0408 | Val Loss: 0.0724\n", "Epoch 93/100 | Train Loss: 0.0406 | Val Loss: 0.0723\n", "Epoch 94/100 | Train Loss: 0.0403 | Val Loss: 0.0725\n", "Epoch 95/100 | Train Loss: 0.0405 | Val Loss: 0.0722\n", "Epoch 96/100 | Train Loss: 0.0408 | Val Loss: 0.0725\n", "Epoch 97/100 | Train Loss: 0.0400 | Val Loss: 0.0727\n", "Epoch 98/100 | Train Loss: 0.0402 | Val Loss: 0.0725\n", "Epoch 99/100 | Train Loss: 0.0420 | Val Loss: 0.0725\n", "Epoch 100/100 | Train Loss: 0.0400 | Val Loss: 0.0726\n" ] } ], "source": [ "# Training loop\n", "for epoch in range(params[\"num_epochs\"]):\n", " model.train()\n", " total_loss = 0\n", " for X_batch, y_batch in train_loader:\n", " optimizer.zero_grad()\n", " y_pred = model(X_batch).squeeze()\n", " loss = criterion(y_pred, y_batch)\n", " loss.backward()\n", " optimizer.step()\n", " total_loss += loss.item()\n", "\n", " avg_train_loss = total_loss / len(train_loader)\n", "\n", " # Validation\n", " model.eval()\n", " val_loss = 0\n", " with torch.no_grad():\n", " for X_val_batch, y_val_batch in val_loader:\n", " y_val_pred = model(X_val_batch).squeeze()\n", " loss = criterion(y_val_pred, y_val_batch)\n", " val_loss += loss.item()\n", "\n", " avg_val_loss = val_loss / len(val_loader)\n", " #scheduler.step(avg_val_loss)\n", " #scheduler.step(val_loss)\n", "\n", "\n", " print(f\"Epoch {epoch+1}/{params['num_epochs']} | \"\n", " f\"Train Loss: {avg_train_loss:.4f} | Val Loss: {avg_val_loss:.4f}\")\n", "\n" ] }, { "cell_type": "code", "execution_count": 875, "id": "ad81cbd3-af54-400a-b39d-45eb56d0113b", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ " Quarter CO_ID Predicted Actual TASI Return Relative_Return\n", "0 2016-01-01 1201.0 -0.029949 -0.196482 0.053471 0.243384\n", "1 2016-04-01 1201.0 -0.025743 0.102171 0.080545 -0.138383\n", "2 2016-07-01 1201.0 -0.021379 -0.295357 -0.081079 -0.102275\n", "3 2016-10-01 1201.0 -0.019016 0.045653 -0.263944 0.087247\n", "4 2017-01-01 1201.0 -0.019200 -0.032564 0.037782 -0.196482\n" ] } ], "source": [ "# Step 1: Make predictions\n", "model.eval()\n", "with torch.no_grad():\n", " predictions = model(X_tensor).squeeze().numpy()\n", "# Step 2: Rebuild the filtered DataFrame to match y_labels\n", "df_sorted = df.sort_values(by=[\"CO_ID\", \"Quarter\"]).reset_index(drop=True)\n", "\n", "# You MUST use the same window_size you used earlier (e.g., 4 or 8)\n", "window_size = 4 # <--- change this if you're using a different one\n", "\n", "# Rebuild the filtered rows (the target rows used in y_labels)\n", "filtered_rows = []\n", "for co_id, group in df_sorted.groupby(\"CO_ID\"):\n", " group = group.reset_index(drop=True)\n", " for i in range(len(group) - window_size):\n", " filtered_rows.append(group.loc[i + window_size])\n", "\n", "df_filtered = pd.DataFrame(filtered_rows).reset_index(drop=True)\n", "\n", "# Step 3: Attach predictions and actuals\n", "df_filtered[\"Predicted\"] = predictions\n", "df_filtered[\"Actual\"] = y_labels\n", "\n", "# Step 4: Attach TASI Return and Relative_Return from df_sorted using index alignment\n", "df_filtered[\"TASI Return\"] = df_sorted.loc[df_filtered.index, \"TASI Return\"].values\n", "df_filtered[\"Relative_Return\"] = df_sorted.loc[df_filtered.index, \"Relative_Return\"].values\n", "\n", "# Step 5: Preview\n", "print(df_filtered[[\"Quarter\", \"CO_ID\", \"Predicted\", \"Actual\", \"TASI Return\", \"Relative_Return\"]].head())\n", "\n" ] }, { "cell_type": "code", "execution_count": 876, "id": "27d1d4d3-c298-46fa-a880-7864f7ee0575", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "['CO_ID', 'Name', 'Year', 'Quarter', 'Sector', 'Return on Assets', 'Return on Capital', 'Return on Equity', 'Return on Common Equity', 'Gross Profit Margin', 'EBITA Margin', 'EBIT Margin', 'Net Income Margin', 'Asset Turnover', 'Current Ratio', 'Debt/Equity', 'Long Term Debt/Captial', 'Book Value Per Share', 'Start Price', 'End Price', 'Stock Return', 'TASI Return', 'Relative_Return']\n" ] } ], "source": [ "print(df_sorted.columns.tolist()) # Check what columns are available\n", "\n" ] }, { "cell_type": "code", "execution_count": 877, "id": "895dd1a3-0e22-483f-a23b-a29fe03aa132", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "MAE: 0.148877\n" ] } ], "source": [ "mae = np.mean(np.abs(df_filtered[\"Predicted\"] - df_filtered[\"Actual\"]))\n", "print(f\"MAE: {mae:.6f}\")" ] }, { "cell_type": "code", "execution_count": 938, "id": "3cbd6f57-b23a-425e-bbaf-b2f556394141", "metadata": {}, "outputs": [], "source": [ "def evaluate_excess_return(df_filtered, k):\n", " # Group by quarter\n", " grouped = df_filtered.groupby(\"Quarter\")\n", "\n", " portfolio_returns = []\n", " benchmark_returns = []\n", "\n", " for quarter, group in grouped:\n", " # Step 1: Select top-k by Predicted return\n", " topk = group.sort_values(by=\"Predicted\", ascending=False).head(k)\n", "\n", " # Step 2: Portfolio return = mean of Actual\n", " R_p_q = topk[\"Actual\"].mean()\n", "\n", " # Step 3: Benchmark return = mean TASI return in this quarter (same for all companies)\n", " R_b_q = group[\"TASI Return\"].iloc[0] # assuming same value per quarter\n", "\n", " portfolio_returns.append(1 + R_p_q)\n", " benchmark_returns.append(1 + R_b_q)\n", "\n", " # Step 4: Multiply all (geometric return)\n", " from functools import reduce\n", " import operator\n", "\n", " def product(lst):\n", " return reduce(operator.mul, lst, 1.0)\n", "\n", " total_portfolio_return = product(portfolio_returns)\n", " total_benchmark_return = product(benchmark_returns)\n", "\n", " excess_return = total_portfolio_return - total_benchmark_return\n", " return excess_return\n" ] }, { "cell_type": "code", "execution_count": 941, "id": "b6561f58-1ca4-4a6c-a76b-a7245add8331", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Excess Return (Top-3): 7.8137\n", "Excess Return (Top-5): 3.3675\n" ] } ], "source": [ "excess_ret = evaluate_excess_return(df_filtered, k=3)\n", "print(f\"Excess Return (Top-3): {excess_ret:.4f}\")\n", "excess_ret = evaluate_excess_return(df_filtered, k=5)\n", "print(f\"Excess Return (Top-5): {excess_ret:.4f}\")" ] }, { "cell_type": "code", "execution_count": 925, "id": "2158a381-cf62-4f5f-8e0c-3543821c9955", "metadata": {}, "outputs": [], "source": [ "def evaluate_excess_return(df_filtered, k=5):\n", " # Group by quarter\n", " grouped = df_filtered.groupby(\"Quarter\")\n", "\n", " portfolio_returns = []\n", " benchmark_returns = []\n", "\n", " for quarter, group in grouped:\n", " # Step 1: Select top-k by Predicted return\n", " topk = group.sort_values(by=\"Predicted\", ascending=False).head(k)\n", "\n", " # Step 2: Portfolio return = mean of Actual\n", " R_p_q = topk[\"Actual\"].mean()\n", "\n", " # Step 3: Benchmark return = mean TASI return in this quarter (same for all companies)\n", " R_b_q = group[\"TASI Return\"].iloc[0] # assuming same value per quarter\n", "\n", " portfolio_returns.append(1 + R_p_q)\n", " benchmark_returns.append(1 + R_b_q)\n", "\n", " # Step 4: Multiply all (geometric return)\n", " from functools import reduce\n", " import operator\n", "\n", " def product(lst):\n", " return reduce(operator.mul, lst, 1.0)\n", "\n", " total_portfolio_return = product(portfolio_returns)\n", " total_benchmark_return = product(benchmark_returns)\n", "\n", " excess_return = total_portfolio_return - total_benchmark_return\n", " return excess_return" ] }, { "cell_type": "code", "execution_count": 926, "id": "85f4bd96-8d5f-4f78-9e9a-9f4d9adee2ee", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Excess Return (Top-20): 3.3675\n" ] } ], "source": [ "excess_ret = evaluate_excess_return(df_filtered, k=5)\n", "print(f\"Excess Return (Top-20): {excess_ret:.4f}\")" ] }, { "cell_type": "code", "execution_count": 927, "id": "a12adc58-0c9a-4624-af04-60ef369e51ba", "metadata": {}, "outputs": [], "source": [ "def evaluate_topk_precision(df_filtered, k=3):\n", " grouped = df_filtered.groupby(\"Quarter\")\n", " precision_scores = []\n", "\n", " for quarter, group in grouped:\n", " topk_pred = set(group.sort_values(\"Predicted\", ascending=False).head(k)[\"CO_ID\"])\n", " topk_actual = set(group.sort_values(\"Actual\", ascending=False).head(k)[\"CO_ID\"])\n", " \n", " intersection = topk_pred & topk_actual\n", " precision = len(intersection) / k\n", " precision_scores.append(precision)\n", "\n", " return np.mean(precision_scores)\n" ] }, { "cell_type": "code", "execution_count": 928, "id": "dfd8a411-26db-4d8a-86a7-7c1053c68380", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Top-k Precision (Top-10): 0.1441\n" ] } ], "source": [ "topk_prec = evaluate_topk_precision(df_filtered, k=3)\n", "print(f\"Top-k Precision (Top-10): {topk_prec:.4f}\")\n" ] }, { "cell_type": "code", "execution_count": 942, "id": "de88b386-e578-4892-896d-7b61faa780c8", "metadata": {}, "outputs": [], "source": [ "def evaluate_topk_precision(df_filtered, k=5):\n", " grouped = df_filtered.groupby(\"Quarter\")\n", " precision_scores = []\n", "\n", " for quarter, group in grouped:\n", " topk_pred = set(group.sort_values(\"Predicted\", ascending=False).head(k)[\"CO_ID\"])\n", " topk_actual = set(group.sort_values(\"Actual\", ascending=False).head(k)[\"CO_ID\"])\n", " \n", " intersection = topk_pred & topk_actual\n", " precision = len(intersection) / k\n", " precision_scores.append(precision)\n", "\n", " return np.mean(precision_scores)" ] }, { "cell_type": "code", "execution_count": 943, "id": "87219fd5-5e7e-48c6-8bf5-627f74259083", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Top-k Precision (Top-10): 0.2216\n" ] } ], "source": [ "topk_prec = evaluate_topk_precision(df_filtered, k=5)\n", "print(f\"Top-k Precision (Top-10): {topk_prec:.4f}\")" ] }, { "cell_type": "code", "execution_count": 931, "id": "90576ea9-ca1c-438e-a178-145b890bdece", "metadata": {}, "outputs": [], "source": [ "def evaluate_portfolio_score(df_filtered, k):\n", " grouped = df_filtered.groupby(\"Quarter\")\n", " returns = []\n", "\n", " for quarter, group in grouped:\n", " topk = group.sort_values(by=\"Predicted\", ascending=False).head(k)\n", " R_p_q = topk[\"Relative_Return\"].mean() # Use relative return per paper\n", " returns.append(R_p_q)\n", "\n", " mean_return = np.mean(returns)\n", " std_return = np.std(returns)\n", "\n", " portfolio_score = mean_return / std_return if std_return != 0 else 0\n", " return portfolio_score\n" ] }, { "cell_type": "code", "execution_count": 933, "id": "e465eb03-a718-4b04-80d2-db758491eec9", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Portfolio Score (Top-10): -0.2738\n", "Portfolio Score (Top-20): -0.4848\n" ] } ], "source": [ "score_10 = evaluate_portfolio_score(df_filtered, k=3)\n", "score_20 = evaluate_portfolio_score(df_filtered, k=5)\n", "\n", "print(f\"Portfolio Score (Top-10): {score_10:.4f}\")\n", "print(f\"Portfolio Score (Top-20): {score_20:.4f}\")\n" ] }, { "cell_type": "code", "execution_count": 888, "id": "9544b2c4-6c07-4fe3-be20-a72f4cf90cbb", "metadata": {}, "outputs": [ { "data": { "image/png": "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", "text/plain": [ "
" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "import matplotlib.pyplot as plt\n", "\n", "# Check the distribution of all relative returns\n", "df_filtered['Relative_Return'].hist(bins=50)\n", "plt.title(\"Distribution of Relative Return\")\n", "plt.xlabel(\"Relative_Return\")\n", "plt.ylabel(\"Frequency\")\n", "plt.show()\n" ] }, { "cell_type": "code", "execution_count": 889, "id": "8f286563-f920-48b2-ae3b-e8b290136a77", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "2016-01-01 00:00:00: Overlap in Top-10 = 1/10\n", "2016-04-01 00:00:00: Overlap in Top-10 = 0/10\n", "2016-07-01 00:00:00: Overlap in Top-10 = 3/10\n", "2016-10-01 00:00:00: Overlap in Top-10 = 1/10\n", "2017-01-01 00:00:00: Overlap in Top-10 = 3/10\n", "2017-04-01 00:00:00: Overlap in Top-10 = 0/10\n", "2017-07-01 00:00:00: Overlap in Top-10 = 1/10\n", "2017-10-01 00:00:00: Overlap in Top-10 = 2/10\n", "2018-01-01 00:00:00: Overlap in Top-10 = 2/10\n", "2018-04-01 00:00:00: Overlap in Top-10 = 1/10\n", "2018-07-01 00:00:00: Overlap in Top-10 = 0/10\n", "2018-10-01 00:00:00: Overlap in Top-10 = 0/10\n", "2019-01-01 00:00:00: Overlap in Top-10 = 1/10\n", "2019-04-01 00:00:00: Overlap in Top-10 = 0/10\n", "2019-07-01 00:00:00: Overlap in Top-10 = 2/10\n", "2019-10-01 00:00:00: Overlap in Top-10 = 0/10\n", "2020-01-01 00:00:00: Overlap in Top-10 = 4/10\n", "2020-04-01 00:00:00: Overlap in Top-10 = 3/10\n", "2020-07-01 00:00:00: Overlap in Top-10 = 5/10\n", "2020-10-01 00:00:00: Overlap in Top-10 = 3/10\n", "2021-01-01 00:00:00: Overlap in Top-10 = 0/10\n", "2021-04-01 00:00:00: Overlap in Top-10 = 2/10\n", "2021-07-01 00:00:00: Overlap in Top-10 = 1/10\n", "2021-10-01 00:00:00: Overlap in Top-10 = 3/10\n", "2022-01-01 00:00:00: Overlap in Top-10 = 2/10\n", "2022-04-01 00:00:00: Overlap in Top-10 = 3/10\n", "2022-07-01 00:00:00: Overlap in Top-10 = 1/10\n", "2022-10-01 00:00:00: Overlap in Top-10 = 2/10\n", "2023-01-01 00:00:00: Overlap in Top-10 = 2/10\n", "2023-04-01 00:00:00: Overlap in Top-10 = 3/10\n", "2023-07-01 00:00:00: Overlap in Top-10 = 1/10\n", "2023-10-01 00:00:00: Overlap in Top-10 = 3/10\n", "2024-01-01 00:00:00: Overlap in Top-10 = 1/10\n", "2024-04-01 00:00:00: Overlap in Top-10 = 1/10\n", "2024-07-01 00:00:00: Overlap in Top-10 = 0/10\n", "2024-10-01 00:00:00: Overlap in Top-10 = 2/10\n", "2025-01-01 00:00:00: Overlap in Top-10 = 0/10\n" ] } ], "source": [ "# Check if predicted top-10 are actually good\n", "for quarter, group in df_filtered.groupby(\"Quarter\"):\n", " topk_pred = group.sort_values(\"Predicted\", ascending=False).head(10)\n", " topk_actual = group.sort_values(\"Relative_Return\", ascending=False).head(10)\n", " overlap = len(set(topk_pred['CO_ID']).intersection(set(topk_actual['CO_ID'])))\n", " print(f\"{quarter}: Overlap in Top-10 = {overlap}/10\")\n" ] }, { "cell_type": "code", "execution_count": 909, "id": "3236bdc5-121e-437e-b98e-da808d872f68", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "MAE (Random Forest): 0.2203\n" ] } ], "source": [ "import pandas as pd\n", "import numpy as np\n", "from sklearn.ensemble import RandomForestRegressor\n", "from sklearn.model_selection import train_test_split\n", "from sklearn.metrics import mean_absolute_error\n", "import matplotlib.pyplot as plt\n", "\n", "# Step 1: Prepare your features and target\n", "# Use the feature columns you've already defined in preprocessing\n", "X = df_filtered[feature_cols] # These should be normalized and cleaned features\n", "y = df_filtered[\"Relative_Return\"]\n", "\n", "# Step 2: Train-test split (no shuffle for time series)\n", "X_train, X_test, y_train, y_test = train_test_split(\n", " X, y, test_size=0.2, shuffle=False\n", ")\n", "\n", "# Step 3: Train the Random Forest model\n", "from sklearn.ensemble import RandomForestRegressor\n", "\n", "rf = RandomForestRegressor(\n", " n_estimators=1000, # Use 1000 trees\n", " criterion='squared_error', # Use squared error as in paper\n", " random_state=42, # For reproducibility\n", " n_jobs=-1 # Use all cores for speed\n", ")\n", "\n", "rf.fit(X_train, y_train)\n", "\n", "# Step 4: Predict on the test set\n", "y_pred = rf.predict(X_test)\n", "\n", "# Step 5: Evaluate with MAE\n", "mae = mean_absolute_error(y_test, y_pred)\n", "print(f\"MAE (Random Forest): {mae:.4f}\")\n", "\n", "# Step 6: Save predictions back to df_filtered\n", "df_filtered.loc[X_test.index, \"Predicted\"] = y_pred\n", "\n", "# Step 7 (Optional): Feature importance plot\n", "importances = rf.feature_importances_\n", "feat_names = X.columns\n", "indices = np.argsort(importances)[::-1]\n", "\n" ] }, { "cell_type": "code", "execution_count": 910, "id": "cb443061-6e0e-4333-91d5-9e7c4e7a129a", "metadata": {}, "outputs": [ { "data": { "image/png": "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", "text/plain": [ "
" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "import matplotlib.pyplot as plt\n", "# Sort features by importance\n", "sorted_idx = importances.argsort()[::-1]\n", "sorted_features = feature_names[sorted_idx]\n", "sorted_importances = importances[sorted_idx]\n", "\n", "# Plot sorted importances\n", "plt.figure(figsize=(8, 6))\n", "plt.bar(x=sorted_features, height=sorted_importances, width=0.6)\n", "plt.title(\"Feature Importance\")\n", "plt.xticks(rotation=90)\n", "plt.ylabel(\"Importance\")\n", "plt.ylim(0, 0.10)\n", "plt.tight_layout()\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": 915, "id": "07f8976d-307c-43e6-b551-f65ddbfc6bb3", "metadata": {}, "outputs": [], "source": [ "# df_test should be created after predicting with RF\n", "df_test = X_test.copy()\n", "df_test['Predicted'] = y_pred\n", "df_test['Relative_Return'] = y_test.values\n", "df_test['Quarter'] = df_filtered.loc[df_test.index, 'Quarter']\n", "df_test['CO_ID'] = df_filtered.loc[df_test.index, 'CO_ID']\n" ] }, { "cell_type": "code", "execution_count": 916, "id": "474c03f8-fdf4-4237-8f31-79aa1c3a128d", "metadata": {}, "outputs": [], "source": [ "def evaluate_portfolio_score(df, k=10):\n", " returns = []\n", " for _, group in df.groupby(\"Quarter\"):\n", " topk = group.sort_values(\"Predicted\", ascending=False).head(k)\n", " r_p_q = topk[\"Relative_Return\"].mean()\n", " returns.append(r_p_q)\n", "\n", " mean_return = np.mean(returns)\n", " std_return = np.std(returns)\n", "\n", " return mean_return / std_return if std_return != 0 else 0\n" ] }, { "cell_type": "code", "execution_count": 917, "id": "d9e1d243-462b-4758-9ca6-19349e3106a5", "metadata": {}, "outputs": [], "source": [ "def evaluate_topk_precision(df, k=10):\n", " correct = []\n", " for _, group in df.groupby(\"Quarter\"):\n", " topk_pred = group.sort_values(\"Predicted\", ascending=False).head(k)['CO_ID']\n", " topk_actual = group.sort_values(\"Relative_Return\", ascending=False).head(k)['CO_ID']\n", " overlap = len(set(topk_pred).intersection(set(topk_actual)))\n", " correct.append(overlap / k)\n", " return np.mean(correct)\n" ] }, { "cell_type": "code", "execution_count": 918, "id": "465a9cb7-7a23-4ac1-9dbb-083aae3d2e61", "metadata": {}, "outputs": [], "source": [ "def evaluate_excess_return(df, k=10):\n", " excess_returns = []\n", " for _, group in df.groupby(\"Quarter\"):\n", " topk = group.sort_values(\"Predicted\", ascending=False).head(k)\n", " r_p_q = topk[\"Relative_Return\"].mean()\n", " r_m_q = group[\"Relative_Return\"].mean()\n", " excess_returns.append(r_p_q - r_m_q)\n", " return np.mean(excess_returns)\n" ] }, { "cell_type": "code", "execution_count": 919, "id": "7b77ab7b-d5a8-4abe-b1b4-8176f52cba0f", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Portfolio Score (Top-10): -0.5149\n", "Top-10 Precision: 1.0000\n", "Top-10 Excess Return: -0.0000\n" ] } ], "source": [ "k = 10\n", "\n", "score = evaluate_portfolio_score(df_test, k)\n", "precision = evaluate_topk_precision(df_test, k)\n", "excess = evaluate_excess_return(df_test, k)\n", "\n", "print(f\"Portfolio Score (Top-{k}): {score:.4f}\")\n", "print(f\"Top-{k} Precision: {precision:.4f}\")\n", "print(f\"Top-{k} Excess Return: {excess:.4f}\")\n" ] }, { "cell_type": "code", "execution_count": 921, "id": "606b0574-8515-4ce6-98cc-4ed71e0637b7", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "RMSE: 0.3639\n" ] } ], "source": [ "from sklearn.ensemble import RandomForestRegressor\n", "from sklearn.metrics import mean_squared_error\n", "import pandas as pd\n", "import numpy as np\n", "\n", "# STEP 1: Top 10 important features from your list (in order of importance)\n", "top_10_features = [\n", " 'Current Ratio',\n", " 'Book Value Per Share',\n", " 'Return on Assets',\n", " 'Return on Capital',\n", " 'Return on Equity',\n", " 'Debt/Equity',\n", " 'Long Term Debt/Captial',\n", " 'Gross Profit Margin',\n", " 'Asset Turnover',\n", " 'Return on Common Equity'\n", "]\n", "\n", "# STEP 2: Filter train and test sets to keep only the top 10 features\n", "X_train_top = X_train[top_10_features].copy()\n", "X_test_top = X_test[top_10_features].copy()\n", "\n", "# STEP 3: Train the Random Forest model\n", "rf_model = RandomForestRegressor(\n", " n_estimators=1000, # as used in your paper\n", " criterion='squared_error', # for regression\n", " random_state=42,\n", " n_jobs=-1\n", ")\n", "rf_model.fit(X_train_top, y_train)\n", "\n", "# STEP 4: Predict on test set\n", "y_pred = rf_model.predict(X_test_top)\n", "\n", "# Optional: Evaluate performance\n", "rmse = np.sqrt(mean_squared_error(y_test, y_pred))\n", "print(f\"RMSE: {rmse:.4f}\")\n", "\n", "# STEP 5: Add predictions back to test dataframe\n", "df_test = X_test_top.copy()\n", "df_test['Predicted'] = y_pred\n", "df_test['Relative_Return'] = y_test.values\n", "df_test['Quarter'] = df_filtered.loc[df_test.index, 'Quarter']\n", "df_test['CO_ID'] = df_filtered.loc[df_test.index, 'CO_ID']\n", "\n" ] }, { "cell_type": "code", "execution_count": null, "id": "87305d35-25d2-486a-a599-2036e80f4a4a", "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "Python 3 (ipykernel)", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.12.9" } }, "nbformat": 4, "nbformat_minor": 5 }