{ "cells": [ { "cell_type": "code", "execution_count": 1, "id": "58890f49-db49-46ca-b999-c4195db4cafe", "metadata": {}, "outputs": [], "source": [ "import pandas as pd\n", "import os\n", "import warnings\n", "warnings.filterwarnings('ignore')" ] }, { "cell_type": "code", "execution_count": 2, "id": "6c2a06ef-97dc-43e3-aa9a-5feb7a2ad3ec", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "C:\\Users\\lirse\\OneDrive\\Desktop\\scripts\\clv\n" ] } ], "source": [ "print(os.getcwd())" ] }, { "cell_type": "code", "execution_count": 3, "id": "5844493a-806f-4c88-804d-3739283e375f", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "['.ipynb_checkpoints', 'clv_research.ipynb', 'live_data', 'optimal_time.csv', 'optimal_time_analyzer.py', 'research.ipynb', 'strategy.ipynb', 'strategy_bets.csv', 'strategy_equity.html', 'strategy_gate_equity.html', 'test.csv', 'test.html', 'test_0.html', 'test_1.html', '__pycache__']\n" ] } ], "source": [ "print(os.listdir(os.getcwd()))" ] }, { "cell_type": "code", "execution_count": 4, "id": "eddbcc72-660f-4235-a54b-b64c97b38f59", "metadata": {}, "outputs": [], "source": [ "dfs = []\n", "for f in os.listdir(os.getcwd()+'/live_data/'):\n", " if '.csv' in f and f != 'test.csv':\n", " df = pd.read_csv(os.getcwd()+'/live_data/'+f)\n", " dfs.append(df)" ] }, { "cell_type": "code", "execution_count": 5, "id": "2129aade-5e95-49f9-802b-0d52d3aa9bed", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "Index(['timestamp', 'game_date', 'league', 'home', 'away', 'status', 'elapsed',\n", " 'home_goals', 'away_goals', 'pin_home', 'pin_draw', 'pin_away',\n", " 'b365_home', 'b365_draw', 'b365_away', 'kalshi_home_mid',\n", " 'kalshi_home_bid', 'kalshi_home_ask', 'kalshi_draw_mid',\n", " 'kalshi_draw_bid', 'kalshi_draw_ask', 'kalshi_away_mid',\n", " 'kalshi_away_bid', 'kalshi_away_ask', 'pin_fair_home', 'pin_fair_draw',\n", " 'pin_fair_away', 'clv_b365_home', 'clv_b365_draw', 'clv_b365_away',\n", " 'clv_kalshi_home', 'clv_kalshi_draw', 'clv_kalshi_away', 'ftr',\n", " 'result_home', 'result_away'],\n", " dtype='object')" ] }, "execution_count": 5, "metadata": {}, "output_type": "execute_result" } ], "source": [ "df = pd.concat(dfs)\n", "df.columns" ] }, { "cell_type": "code", "execution_count": 6, "id": "12be0c27-8d6a-42ab-a452-bf7d5ede7c58", "metadata": {}, "outputs": [], "source": [ "def highest_clv(home, draw, away):\n", " clv_dict = {'H': home, 'D': draw, 'A': away}\n", " max_clv = None if pd.isna(clv_dict['H']) else max(clv_dict, key=clv_dict.get)\n", " return max_clv, max(clv_dict.values())\n", "\n", "bet_df = df\n", "bet_df['b365_highest_clv'] = bet_df.apply(lambda row: highest_clv(row['clv_b365_home'], row['clv_b365_draw'], row['clv_b365_away'])[0], axis=1)\n", "bet_df['b365_highest_clv_val'] = bet_df.apply(lambda row: highest_clv(row['clv_b365_home'], row['clv_b365_draw'], row['clv_b365_away'])[1], axis=1)\n", "bet_df['kalshi_highest_clv'] = bet_df.apply(lambda row: highest_clv(row['clv_kalshi_home'], row['clv_kalshi_draw'], row['clv_kalshi_away'])[0], axis=1)\n", "bet_df['kalshi_highest_clv_val'] = bet_df.apply(lambda row: highest_clv(row['clv_kalshi_home'], row['clv_kalshi_draw'], row['clv_kalshi_away'])[1], axis=1)" ] }, { "cell_type": "code", "execution_count": 7, "id": "e976a383-9710-4e67-8687-d69d27601411", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "Index(['timestamp', 'game_date', 'league', 'home', 'away', 'status', 'elapsed',\n", " 'home_goals', 'away_goals', 'pin_home', 'pin_draw', 'pin_away',\n", " 'b365_home', 'b365_draw', 'b365_away', 'kalshi_home_mid',\n", " 'kalshi_home_bid', 'kalshi_home_ask', 'kalshi_draw_mid',\n", " 'kalshi_draw_bid', 'kalshi_draw_ask', 'kalshi_away_mid',\n", " 'kalshi_away_bid', 'kalshi_away_ask', 'pin_fair_home', 'pin_fair_draw',\n", " 'pin_fair_away', 'clv_b365_home', 'clv_b365_draw', 'clv_b365_away',\n", " 'clv_kalshi_home', 'clv_kalshi_draw', 'clv_kalshi_away', 'ftr',\n", " 'result_home', 'result_away', 'b365_highest_clv',\n", " 'b365_highest_clv_val', 'kalshi_highest_clv', 'kalshi_highest_clv_val',\n", " 'correct_b365', 'correct_kalshi'],\n", " dtype='object')" ] }, "execution_count": 7, "metadata": {}, "output_type": "execute_result" } ], "source": [ "bet_df['correct_b365'] = bet_df['ftr'] == bet_df['b365_highest_clv']\n", "bet_df['correct_b365'] = bet_df['correct_b365'].apply(lambda x: 1 if x == True else 0)\n", "bet_df['correct_kalshi'] = bet_df['ftr'] == bet_df['kalshi_highest_clv']\n", "bet_df['correct_kalshi'] = bet_df['correct_kalshi'].apply(lambda x: 1 if x == True else 0)\n", "bet_df.columns" ] }, { "cell_type": "code", "execution_count": 8, "id": "32b03913-2884-4f08-8a2b-dbfd1211ef04", "metadata": { "scrolled": true }, "outputs": [], "source": [ "bet_dict = {'H': '_home', 'D': '_draw', 'A': '_away'}\n", "df_filtered_nan = bet_df.dropna(subset=['b365_highest_clv', 'kalshi_highest_clv'])\n", "df_filtered_nan['b365_bet'] = df_filtered_nan.apply(lambda row: row['b365'+bet_dict[row['b365_highest_clv']]], axis=1)\n", "df_filtered_nan['kalshi_bet'] = df_filtered_nan.apply(lambda row: row['kalshi'+bet_dict[row['kalshi_highest_clv']]+'_ask'], axis=1)\n", "df_filtered_nan['kalshi_bet_b365'] = df_filtered_nan.apply(lambda row: row['kalshi'+bet_dict[row['b365_highest_clv']]+'_ask'], axis=1)" ] }, { "cell_type": "code", "execution_count": 9, "id": "ff62d5b4-e1f2-4cfa-a2fd-53a32afdaea7", "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", " \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", "
timestampgame_dateleaguehomeawaystatuselapsedhome_goalsaway_goalspin_home...result_awayb365_highest_clvb365_highest_clv_valkalshi_highest_clvkalshi_highest_clv_valcorrect_b365correct_kalshib365_betkalshi_betkalshi_bet_b365
04/22/2026 0:012026-04-22E1CharltonIpswichPRE-883.00.00.05.11...2.0A-0.051139H0.041901101.700.190.57
14/22/2026 0:012026-04-22E1BirminghamPrestonPRE-883.00.00.01.67...1.0H-0.075843A0.052148101.620.180.58
24/22/2026 0:012026-04-22E1MiddlesbroughSheffield WednesdayPRE-883.00.00.01.16...0.0H-0.041614A0.191379101.170.060.84
34/22/2026 0:012026-04-22E0BurnleyManchester CityPRE-898.00.00.016.43...1.0H-0.001936D0.0732430017.000.100.06
44/22/2026 0:012026-04-22E0BournemouthLeeds UnitedPRE-898.00.00.02.01...2.0A-0.060761H-0.007949003.700.490.27
..................................................................
2772026-06-14 12:05:382026-06-14SP2MalagaAlmeriaPRE-174.00.00.02.27...NaND-0.037752D0.018964003.200.300.30
2922026-06-14 12:21:012026-06-14CANVancouver FCPacific FCPRE-338.00.00.01.66...NaNA0.000175A0.122654004.500.200.20
2942026-06-14 12:21:022026-06-14CL1Union La CaleraUniversidad de ChilePRE-158.00.00.03.57...NaND-0.011568D0.031800003.300.300.30
2952026-06-14 12:21:022026-06-14CL1U. CatolicaUniversidad de ConcepcionPRE-308.00.00.01.56...NaNH-0.047179A0.139857001.550.140.66
2962026-06-14 12:21:032026-06-14SP2MalagaAlmeriaPRE-158.00.00.02.27...NaND-0.037752A0.028414003.200.270.30
\n", "

14432 rows \u00d7 45 columns

\n", "
" ], "text/plain": [ " timestamp game_date league home \\\n", "0 4/22/2026 0:01 2026-04-22 E1 Charlton \n", "1 4/22/2026 0:01 2026-04-22 E1 Birmingham \n", "2 4/22/2026 0:01 2026-04-22 E1 Middlesbrough \n", "3 4/22/2026 0:01 2026-04-22 E0 Burnley \n", "4 4/22/2026 0:01 2026-04-22 E0 Bournemouth \n", ".. ... ... ... ... \n", "277 2026-06-14 12:05:38 2026-06-14 SP2 Malaga \n", "292 2026-06-14 12:21:01 2026-06-14 CAN Vancouver FC \n", "294 2026-06-14 12:21:02 2026-06-14 CL1 Union La Calera \n", "295 2026-06-14 12:21:02 2026-06-14 CL1 U. Catolica \n", "296 2026-06-14 12:21:03 2026-06-14 SP2 Malaga \n", "\n", " away status elapsed home_goals away_goals \\\n", "0 Ipswich PRE -883.0 0.0 0.0 \n", "1 Preston PRE -883.0 0.0 0.0 \n", "2 Sheffield Wednesday PRE -883.0 0.0 0.0 \n", "3 Manchester City PRE -898.0 0.0 0.0 \n", "4 Leeds United PRE -898.0 0.0 0.0 \n", ".. ... ... ... ... ... \n", "277 Almeria PRE -174.0 0.0 0.0 \n", "292 Pacific FC PRE -338.0 0.0 0.0 \n", "294 Universidad de Chile PRE -158.0 0.0 0.0 \n", "295 Universidad de Concepcion PRE -308.0 0.0 0.0 \n", "296 Almeria PRE -158.0 0.0 0.0 \n", "\n", " pin_home ... result_away b365_highest_clv b365_highest_clv_val \\\n", "0 5.11 ... 2.0 A -0.051139 \n", "1 1.67 ... 1.0 H -0.075843 \n", "2 1.16 ... 0.0 H -0.041614 \n", "3 16.43 ... 1.0 H -0.001936 \n", "4 2.01 ... 2.0 A -0.060761 \n", ".. ... ... ... ... ... \n", "277 2.27 ... NaN D -0.037752 \n", "292 1.66 ... NaN A 0.000175 \n", "294 3.57 ... NaN D -0.011568 \n", "295 1.56 ... NaN H -0.047179 \n", "296 2.27 ... NaN D -0.037752 \n", "\n", " kalshi_highest_clv kalshi_highest_clv_val correct_b365 correct_kalshi \\\n", "0 H 0.041901 1 0 \n", "1 A 0.052148 1 0 \n", "2 A 0.191379 1 0 \n", "3 D 0.073243 0 0 \n", "4 H -0.007949 0 0 \n", ".. ... ... ... ... \n", "277 D 0.018964 0 0 \n", "292 A 0.122654 0 0 \n", "294 D 0.031800 0 0 \n", "295 A 0.139857 0 0 \n", "296 A 0.028414 0 0 \n", "\n", " b365_bet kalshi_bet kalshi_bet_b365 \n", "0 1.70 0.19 0.57 \n", "1 1.62 0.18 0.58 \n", "2 1.17 0.06 0.84 \n", "3 17.00 0.10 0.06 \n", "4 3.70 0.49 0.27 \n", ".. ... ... ... \n", "277 3.20 0.30 0.30 \n", "292 4.50 0.20 0.20 \n", "294 3.30 0.30 0.30 \n", "295 1.55 0.14 0.66 \n", "296 3.20 0.27 0.30 \n", "\n", "[14432 rows x 45 columns]" ] }, "execution_count": 9, "metadata": {}, "output_type": "execute_result" } ], "source": [ "df_filtered_nan" ] }, { "cell_type": "code", "execution_count": 10, "id": "0bde1df3-2da1-417c-b0f0-9cedde4813f6", "metadata": {}, "outputs": [], "source": [ "df_filtered_nan['returns_b365'] = df_filtered_nan.apply(lambda row: row['b365_bet'] if row['correct_b365'] == 1 else -1, axis=1)\n", "df_filtered_nan['returns_kalshi'] = df_filtered_nan.apply(lambda row: row['kalshi_bet'] if row['correct_kalshi'] == 1 else -1, axis=1)\n", "df_filtered_nan['returns_kalshi_b365'] = df_filtered_nan.apply(lambda row: row['kalshi_bet_b365'] if row['correct_b365'] == 1 else -1, axis=1)" ] }, { "cell_type": "code", "execution_count": 11, "id": "89e00959-acca-4322-9226-2b01ac5b097d", "metadata": {}, "outputs": [], "source": [ "df_filtered_nan['returns_kalshi_odds'] = 1/df_filtered_nan['returns_kalshi']\n", "df_filtered_nan['returns_kalshi_b365_odds'] = 1/df_filtered_nan['returns_kalshi_b365']" ] }, { "cell_type": "code", "execution_count": 12, "id": "45c06182-ad8b-438d-b589-bc20be205c95", "metadata": {}, "outputs": [], "source": [ "# Pick one snapshot per game: closest to TARGET_MINS before kickoff\n", "TARGET_MINS = 480 # 8 hours before kickoff\n", "df_filtered_nan['dist_to_target'] = (df_filtered_nan['elapsed'] - (-TARGET_MINS)).abs()\n", "df_filtered_nan = df_filtered_nan.sort_values('dist_to_target')\n", "df_filtered_nan = df_filtered_nan.drop_duplicates(subset=['home', 'away'], keep='first')\n", "df_filtered_nan = df_filtered_nan.drop(columns='dist_to_target')\n", "\n", "# Only bet when the chosen side actually has positive CLV\n", "clv_map = {'H': 'clv_b365_home', 'D': 'clv_b365_draw', 'A': 'clv_b365_away'}\n", "df_filtered_nan['max_b365_clv'] = df_filtered_nan.apply(lambda row: row[clv_map[row['b365_highest_clv']]], axis=1)\n", "df_filtered_nan = df_filtered_nan[df_filtered_nan['max_b365_clv'] > 0]\n", "\n", "df_filtered_nan = df_filtered_nan[df_filtered_nan['b365_highest_clv']!='D']\n", "df_filtered_nan = df_filtered_nan.dropna(subset=['result_home', 'result_away', 'clv_kalshi_home', 'clv_kalshi_draw', 'clv_kalshi_away'], how='any')" ] }, { "cell_type": "code", "execution_count": 13, "id": "9d96bebe-805b-4bd7-a001-8bcfb0f82e82", "metadata": {}, "outputs": [], "source": [ "def kelly_size(probability, odds, bankroll, max_fraction=0.5):\n", " b = odds - 1 # net odds\n", " f_star = (probability * odds - 1) / b # binary Kelly = (p - ask)/(1 - ask)\n", " f_capped = min(f_star, max_fraction)\n", " return max(0, f_capped * bankroll)" ] }, { "cell_type": "code", "execution_count": 14, "id": "38633e9f-8a8b-4016-abbe-adbf25ce0ea3", "metadata": {}, "outputs": [], "source": [ "def cumulative_kelly(bets, probabilities, outcomes, bankroll=1000, max_fraction=0.5):\n", " \"\"\"\n", " Simulate cumulative Kelly betting over a series of bets.\n", "\n", " Args:\n", " bets: List of decimal odds for each bet\n", " probabilities: List of win probabilities for each bet\n", " outcomes: List of 1 (win) or 0 (loss) for each bet\n", " bankroll: Starting bankroll\n", " max_fraction: Max fraction of bankroll to bet (0.5 = half-Kelly)\n", "\n", " Returns:\n", " List of (bet_amount, new_bankroll) tuples for each step\n", " \"\"\"\n", " results = []\n", " br = bankroll\n", "\n", " for odds, prob, outcome in zip(bets, probabilities, outcomes):\n", " bet = kelly_size(prob, odds, br, max_fraction)\n", " if outcome == 1:\n", " br += bet * (odds - 1)\n", " else:\n", " br -= bet\n", " results.append((bet, br))\n", "\n", " return results" ] }, { "cell_type": "code", "execution_count": 15, "id": "7445c67d-efc3-4c91-b208-abfe01dee292", "metadata": {}, "outputs": [], "source": [ "# Use Pinnacle fair probability as p_win for Kelly (not implied 1/odds)\n", "pin_fair_map = {'H': 'pin_fair_home', 'D': 'pin_fair_draw', 'A': 'pin_fair_away'}\n", "df_filtered_nan['pin_fair_b365'] = df_filtered_nan.apply(lambda row: row[pin_fair_map[row['b365_highest_clv']]], axis=1)\n", "df_filtered_nan['pin_fair_kalshi'] = df_filtered_nan.apply(lambda row: row[pin_fair_map[row['kalshi_highest_clv']]], axis=1)\n", "\n", "kelly_b365 = cumulative_kelly(df_filtered_nan['b365_bet'], df_filtered_nan['pin_fair_b365'], df_filtered_nan['correct_b365'], max_fraction=0.025)\n", "df_filtered_nan['kelly_b365'] = [x[0] for x in kelly_b365]\n", "df_filtered_nan['kelly_b365_comp'] = [x[1] for x in kelly_b365]\n", "kelly_kalshi = cumulative_kelly(1/df_filtered_nan['kalshi_bet'], df_filtered_nan['pin_fair_kalshi'], df_filtered_nan['correct_kalshi'], max_fraction=0.025)\n", "df_filtered_nan['kelly_kalshi'] = [x[0] for x in kelly_kalshi]\n", "df_filtered_nan['kelly_kalshi_comp'] = [x[1] for x in kelly_kalshi]\n", "kelly_kalshi_b365 = cumulative_kelly(1/df_filtered_nan['kalshi_bet_b365'], df_filtered_nan['pin_fair_b365'], df_filtered_nan['correct_b365'], max_fraction=0.025)\n", "df_filtered_nan['kelly_kalshi_b365'] = [x[0] for x in kelly_kalshi_b365]\n", "df_filtered_nan['kelly_kalshi_b365_comp'] = [x[1] for x in kelly_kalshi_b365]" ] }, { "cell_type": "code", "execution_count": 16, "id": "0913c2ad-dfbb-4870-a552-2eb73b0d1ead", "metadata": {}, "outputs": [], "source": [ "import plotly.express as px\n", "\n", "x = range(df_filtered_nan.shape[0])\n", "y = df_filtered_nan['kelly_kalshi_b365_comp']\n", "\n", "fig = px.line(x=x, y=y, title='Compounded Returns', labels={'x':'Bets', 'y':'Returns'})\n", "fig.write_html('test_0.html')" ] }, { "cell_type": "code", "execution_count": 17, "id": "833af46b-1f8d-4a0c-9dbd-7d834e77152e", "metadata": {}, "outputs": [], "source": [ "# df_filtered_nan.to_csv('test.csv', index=False)" ] }, { "cell_type": "code", "execution_count": 18, "id": "de201731-df9b-4d80-bc96-3a1533020189", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Summary\n", "==========================================================================================\n", "Total Bets: 26\n", "Correct B365: 0.11538461538461539\n", "Correct Kalshi: 0.07692307692307693\n", "Flat B365 Returns: $-8.57\n", "Flat Kalshi Returns: $-23.62\n", "Flat Kalshi w/ B365 Basis Return: $-8.99\n", "Final B365 Return (Compounded): $975.21\n", "Final Kalshi Return (Compounded): $833.64\n", "Final Kalshi w/ B365 Basis Return (Compounded): $887.81\n" ] } ], "source": [ "print('Summary')\n", "print('==========================================================================================')\n", "print(f'Total Bets: {df_filtered_nan.shape[0]}')\n", "print(f'Correct B365: {df_filtered_nan[\"correct_b365\"].mean()}')\n", "print(f'Correct Kalshi: {df_filtered_nan[\"correct_kalshi\"].mean()}')\n", "print(f'Flat B365 Returns: ${df_filtered_nan[\"returns_b365\"].sum():,.2f}')\n", "print(f'Flat Kalshi Returns: ${df_filtered_nan[\"returns_kalshi\"].sum():,.2f}')\n", "print(f'Flat Kalshi w/ B365 Basis Return: ${df_filtered_nan[\"returns_kalshi_b365_odds\"].sum():,.2f}')\n", "print(f'Final B365 Return (Compounded): ${df_filtered_nan[\"kelly_b365_comp\"].iloc[-1]:,.2f}')\n", "print(f'Final Kalshi Return (Compounded): ${df_filtered_nan[\"kelly_kalshi_comp\"].iloc[-1]:,.2f}')\n", "print(f'Final Kalshi w/ B365 Basis Return (Compounded): ${df_filtered_nan[\"kelly_kalshi_b365_comp\"].iloc[-1]:,.2f}')\n" ] }, { "cell_type": "code", "execution_count": 19, "id": "0f0427c0-5446-436d-b53e-49f84bb44015", "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", " \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", "
timestampgame_dateleaguehomeawaystatuselapsedhome_goalsaway_goalspin_home...clv_kalshi_awayftrresult_homeresult_awayb365_highest_clvb365_highest_clv_valkalshi_highest_clvkalshi_highest_clv_valcorrect_b365correct_kalshi
04/22/2026 0:012026-04-22E1CharltonIpswichPRE-883.00.00.05.11...-0.012265A1.02.0A-0.051139H0.04190110
14/22/2026 0:012026-04-22E1BirminghamPrestonPRE-883.00.00.01.67...0.052148H2.01.0H-0.075843A0.05214810
24/22/2026 0:012026-04-22E1MiddlesbroughSheffield WednesdayPRE-883.00.00.01.16...0.191379H1.00.0H-0.041614A0.19137910
34/22/2026 0:012026-04-22E0BurnleyManchester CityPRE-898.00.00.016.43...-0.031257A0.01.0H-0.001936D0.07324300
44/22/2026 0:012026-04-22E0BournemouthLeeds UnitedPRE-898.00.00.02.01...-0.024235D2.02.0A-0.060761H-0.00794900
..................................................................
2772026-06-14 12:05:382026-06-14SP2MalagaAlmeriaPRE-174.00.00.02.27...0.010082NaNNaNNaND-0.037752D0.01896400
2922026-06-14 12:21:012026-06-14CANVancouver FCPacific FCPRE-338.00.00.01.66...0.122654NaNNaNNaNA0.000175A0.12265400
2942026-06-14 12:21:022026-06-14CL1Union La CaleraUniversidad de ChilePRE-158.00.00.03.57...-0.032112NaNNaNNaND-0.011568D0.03180000
2952026-06-14 12:21:022026-06-14CL1U. CatolicaUniversidad de ConcepcionPRE-308.00.00.01.56...0.139857NaNNaNNaNH-0.047179A0.13985700
2962026-06-14 12:21:032026-06-14SP2MalagaAlmeriaPRE-158.00.00.02.27...0.028414NaNNaNNaND-0.037752A0.02841400
\n", "

14432 rows \u00d7 42 columns

\n", "
" ], "text/plain": [ " timestamp game_date league home \\\n", "0 4/22/2026 0:01 2026-04-22 E1 Charlton \n", "1 4/22/2026 0:01 2026-04-22 E1 Birmingham \n", "2 4/22/2026 0:01 2026-04-22 E1 Middlesbrough \n", "3 4/22/2026 0:01 2026-04-22 E0 Burnley \n", "4 4/22/2026 0:01 2026-04-22 E0 Bournemouth \n", ".. ... ... ... ... \n", "277 2026-06-14 12:05:38 2026-06-14 SP2 Malaga \n", "292 2026-06-14 12:21:01 2026-06-14 CAN Vancouver FC \n", "294 2026-06-14 12:21:02 2026-06-14 CL1 Union La Calera \n", "295 2026-06-14 12:21:02 2026-06-14 CL1 U. Catolica \n", "296 2026-06-14 12:21:03 2026-06-14 SP2 Malaga \n", "\n", " away status elapsed home_goals away_goals \\\n", "0 Ipswich PRE -883.0 0.0 0.0 \n", "1 Preston PRE -883.0 0.0 0.0 \n", "2 Sheffield Wednesday PRE -883.0 0.0 0.0 \n", "3 Manchester City PRE -898.0 0.0 0.0 \n", "4 Leeds United PRE -898.0 0.0 0.0 \n", ".. ... ... ... ... ... \n", "277 Almeria PRE -174.0 0.0 0.0 \n", "292 Pacific FC PRE -338.0 0.0 0.0 \n", "294 Universidad de Chile PRE -158.0 0.0 0.0 \n", "295 Universidad de Concepcion PRE -308.0 0.0 0.0 \n", "296 Almeria PRE -158.0 0.0 0.0 \n", "\n", " pin_home ... clv_kalshi_away ftr result_home result_away \\\n", "0 5.11 ... -0.012265 A 1.0 2.0 \n", "1 1.67 ... 0.052148 H 2.0 1.0 \n", "2 1.16 ... 0.191379 H 1.0 0.0 \n", "3 16.43 ... -0.031257 A 0.0 1.0 \n", "4 2.01 ... -0.024235 D 2.0 2.0 \n", ".. ... ... ... ... ... ... \n", "277 2.27 ... 0.010082 NaN NaN NaN \n", "292 1.66 ... 0.122654 NaN NaN NaN \n", "294 3.57 ... -0.032112 NaN NaN NaN \n", "295 1.56 ... 0.139857 NaN NaN NaN \n", "296 2.27 ... 0.028414 NaN NaN NaN \n", "\n", " b365_highest_clv b365_highest_clv_val kalshi_highest_clv \\\n", "0 A -0.051139 H \n", "1 H -0.075843 A \n", "2 H -0.041614 A \n", "3 H -0.001936 D \n", "4 A -0.060761 H \n", ".. ... ... ... \n", "277 D -0.037752 D \n", "292 A 0.000175 A \n", "294 D -0.011568 D \n", "295 H -0.047179 A \n", "296 D -0.037752 A \n", "\n", " kalshi_highest_clv_val correct_b365 correct_kalshi \n", "0 0.041901 1 0 \n", "1 0.052148 1 0 \n", "2 0.191379 1 0 \n", "3 0.073243 0 0 \n", "4 -0.007949 0 0 \n", ".. ... ... ... \n", "277 0.018964 0 0 \n", "292 0.122654 0 0 \n", "294 0.031800 0 0 \n", "295 0.139857 0 0 \n", "296 0.028414 0 0 \n", "\n", "[14432 rows x 42 columns]" ] }, "execution_count": 19, "metadata": {}, "output_type": "execute_result" } ], "source": [ "df_filtered_nan_ = df.dropna(subset=['b365_highest_clv', 'kalshi_highest_clv'])\n", "df_filtered_nan_" ] }, { "cell_type": "markdown", "id": "89409757-e0e0-4f64-b10d-f812d0653993", "metadata": {}, "source": [ "# Time-based analysis" ] }, { "cell_type": "code", "execution_count": 20, "id": "bc4dd89d-756c-423b-83b5-df49a8eb3769", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "array([-1349. , -1129.5, -910. , -690.5, -471. , -251.5, -32. ])" ] }, "execution_count": 20, "metadata": {}, "output_type": "execute_result" } ], "source": [ "import math\n", "import numpy as np\n", "\n", "df_filtered_nan_ = df_filtered_nan_[df_filtered_nan_['status']=='PRE']\n", "num_bins = math.ceil(abs(int(df_filtered_nan_['elapsed'].min()))/240)\n", "bins = np.histogram(df_filtered_nan_['elapsed'], bins=num_bins)[1]\n", "bins" ] }, { "cell_type": "code", "execution_count": 21, "id": "ee953fe7-29f0-415b-be76-4a02e47647ee", "metadata": {}, "outputs": [], "source": [ "def assign_bin(number):\n", " bins = np.histogram(df_filtered_nan_['elapsed'], bins=num_bins)[1]\n", " for i in range(len(bins)+1):\n", " if number >= bins[i] and number <= bins[i+1]:\n", " return i" ] }, { "cell_type": "code", "execution_count": 22, "id": "ed296290-4b77-4f60-87c7-10e1ecff5063", "metadata": {}, "outputs": [], "source": [ "df_filtered_nan_['bin'] = df_filtered_nan_['elapsed'].apply(lambda x: assign_bin(x))" ] }, { "cell_type": "code", "execution_count": 23, "id": "8d10b264-63ea-4185-b463-cac1515c79d5", "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", "
bincorrect_b365correct_kalshibin_rangebin_mid
meancountmeancount
000.202532790.291139791349-11301239
110.4081631960.2551021961130-9101020
220.5042743510.304843351910-690800
330.4041107300.267123730690-471581
440.3972138610.297329861471-252361
550.3886406690.285501669252-32142
\n", "
" ], "text/plain": [ " bin correct_b365 correct_kalshi bin_range bin_mid\n", " mean count mean count \n", "0 0 0.202532 79 0.291139 79 1349-1130 1239\n", "1 1 0.408163 196 0.255102 196 1130-910 1020\n", "2 2 0.504274 351 0.304843 351 910-690 800\n", "3 3 0.404110 730 0.267123 730 690-471 581\n", "4 4 0.397213 861 0.297329 861 471-252 361\n", "5 5 0.388640 669 0.285501 669 252-32 142" ] }, "execution_count": 23, "metadata": {}, "output_type": "execute_result" } ], "source": [ "first_20 = round(df_filtered_nan_.shape[0]*.2)\n", "df_first_20 = df_filtered_nan_.head(first_20)\n", "df_binned = (\n", " df_first_20[['correct_b365', 'correct_kalshi', 'bin']]\n", " .groupby(by='bin')\n", " .agg({'correct_b365': ['mean', 'count'],\n", " 'correct_kalshi': ['mean', 'count'],\n", " # 'b365_highest_clv': ['mean']\n", " })\n", " .reset_index()\n", ")\n", "df_binned['bin_range'] = df_binned['bin'].apply(lambda x: f'{-1*round(bins[x])}-{-1*round(bins[x+1])}')\n", "df_binned['bin_mid'] = df_binned['bin'].apply(lambda x: f'{-1*round((bins[x]+bins[x+1])/2)}')\n", "df_binned" ] }, { "cell_type": "code", "execution_count": 24, "id": "abb4fa36-3731-4e37-9aa9-9f7b22740c24", "metadata": {}, "outputs": [], "source": [ "best_bin = df_binned.sort_values(by=('correct_kalshi', 'mean'), ascending=False)['bin'].iloc[0]\n", "TARGET_MINS = int(df_binned.sort_values(by=('correct_kalshi', 'mean'), ascending=False)['bin_mid'].iloc[0])\n", "df_remaining = df_filtered_nan_[first_20:]\n", "binned_df = df_remaining#[(df_remaining['league']!='USL')&(df_remaining['league']!='BSA')]\n", "binned_df = binned_df[(binned_df['kalshi_highest_clv_val']>0.04)]\n", "binned_df = binned_df[binned_df['elapsed'] >= -TARGET_MINS] # countdown has reached the target\n", "binned_df['dist_to_target'] = binned_df['elapsed'] + TARGET_MINS # = elapsed - (-TARGET_MINS), always >= 0\n", "binned_df = binned_df.sort_values('dist_to_target')\n", "binned_df = binned_df.drop_duplicates(subset=['home', 'away'], keep='first')\n", "binned_df = binned_df.drop(columns='dist_to_target')\n", "binned_df['timestamp'] = binned_df['timestamp'].apply(lambda x: pd.to_datetime(x).timestamp())\n", "binned_df = binned_df.sort_values(by='timestamp')" ] }, { "cell_type": "code", "execution_count": 25, "id": "6e923bb4-37e5-44a9-9e7c-63cbe842f17b", "metadata": {}, "outputs": [], "source": [ "binned_df['b365_bet'] = binned_df.apply(lambda row: row['b365'+bet_dict[row['b365_highest_clv']]], axis=1)\n", "binned_df['kalshi_bet'] = binned_df.apply(lambda row: row['kalshi'+bet_dict[row['kalshi_highest_clv']]+'_ask'], axis=1)\n", "binned_df['kalshi_bet_b365'] = binned_df.apply(lambda row: row['kalshi'+bet_dict[row['b365_highest_clv']]+'_ask'], axis=1)\n", "binned_df['returns_b365'] = binned_df.apply(lambda row: row['b365_bet']-1 if row['correct_b365'] == 1 else -1, axis=1)\n", "binned_df['returns_kalshi'] = binned_df.apply(lambda row: 1/row['kalshi_bet']-1 if row['correct_kalshi'] == 1 else -1, axis=1)\n", "binned_df['returns_kalshi_b365'] = binned_df.apply(lambda row: 1/row['kalshi_bet_b365']-1 if row['correct_b365'] == 1 else -1, axis=1)\n", "# binned_df = binned_df[binned_df['b365_highest_clv']!='D']\n", "binned_df = binned_df.dropna(subset=['result_home', 'result_away', 'clv_kalshi_home', 'clv_kalshi_draw', 'clv_kalshi_away'], how='any')" ] }, { "cell_type": "code", "execution_count": 26, "id": "92d99b27-6e85-4556-b240-202696571476", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "SKIP: 164 BET: 133 BENCH: 135\n" ] } ], "source": [ "from collections import defaultdict\n", "\n", "WINDOW, WARMUP = 10, 8\n", "binned_df = binned_df.sort_values('timestamp').reset_index(drop=True)\n", "\n", "hist = defaultdict(list) # league -> net outcome of every signal seen so far\n", "decisions, rolls = [], []\n", "for _, row in binned_df.iterrows():\n", " h = hist[row['league']]\n", " net = row['returns_kalshi'] # flat net this signal would earn (1/ask-1 if win else -1)\n", " if len(h) < WARMUP:\n", " dec, roll = 'SKIP', np.nan # not enough history -> don't bet yet\n", " else:\n", " roll = np.mean(h[-WINDOW:]) # rolling mean net over last WINDOW signals\n", " dec = 'BET' if roll >= 0 else 'BENCH'\n", " decisions.append(dec); rolls.append(roll)\n", " h.append(net) # record outcome regardless of decision (point-in-time)\n", "\n", "binned_df['league_roll_roi'] = rolls\n", "binned_df['decision'] = decisions\n", "\n", "print('SKIP:', (binned_df.decision=='SKIP').sum(),\n", " 'BET:', (binned_df.decision=='BET').sum(),\n", " 'BENCH:', (binned_df.decision=='BENCH').sum())" ] }, { "cell_type": "code", "execution_count": 27, "id": "cc40ef8d-3799-4245-a526-4a436249cfd8", "metadata": {}, "outputs": [], "source": [ "# Use Pinnacle fair probability as p_win for Kelly\n", "binned_df['pin_fair_b365'] = binned_df.apply(lambda row: row[pin_fair_map[row['b365_highest_clv']]], axis=1)\n", "binned_df['pin_fair_kalshi'] = binned_df.apply(lambda row: row[pin_fair_map[row['kalshi_highest_clv']]], axis=1)\n", "\n", "kelly_b365 = cumulative_kelly(binned_df['b365_bet'], binned_df['pin_fair_b365'], binned_df['correct_b365'], max_fraction=0.025)\n", "binned_df['kelly_b365'] = [x[0] for x in kelly_b365]\n", "binned_df['kelly_b365_comp'] = [x[1] for x in kelly_b365]\n", "kelly_kalshi = cumulative_kelly(1/binned_df['kalshi_bet'], binned_df['pin_fair_kalshi'], binned_df['correct_kalshi'], max_fraction=0.025)\n", "binned_df['kelly_kalshi'] = [x[0] for x in kelly_kalshi]\n", "binned_df['kelly_kalshi_comp'] = [x[1] for x in kelly_kalshi]\n", "kelly_kalshi_b365 = cumulative_kelly(1/binned_df['kalshi_bet_b365'], binned_df['pin_fair_b365'], binned_df['correct_b365'], max_fraction=0.025)\n", "binned_df['kelly_kalshi_b365'] = [x[0] for x in kelly_kalshi_b365]\n", "binned_df['kelly_kalshi_b365_comp'] = [x[1] for x in kelly_kalshi_b365]" ] }, { "cell_type": "code", "execution_count": 28, "id": "77442985-a553-48cd-b339-e15c7e5fd5ae", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "gated Kalshi flat ROI: -13.1% (n=133)\n", "gated Kalshi compounded: $773.17\n" ] } ], "source": [ "bet_df = binned_df[binned_df['decision'] == 'BET'].copy()\n", "k = cumulative_kelly(1/bet_df['kalshi_bet'], bet_df['pin_fair_kalshi'],\n", " bet_df['correct_kalshi'], max_fraction=0.025)\n", "bet_df['kelly_kalshi_comp'] = [x[1] for x in k]\n", "\n", "print(f\"gated Kalshi flat ROI: {100*bet_df['returns_kalshi'].mean():+.1f}% (n={len(bet_df)})\")\n", "print(f\"gated Kalshi compounded: ${bet_df['kelly_kalshi_comp'].iloc[-1]:,.2f}\")" ] }, { "cell_type": "code", "execution_count": 29, "id": "20a89e95-311b-456f-9f14-dfb0aad7b5bf", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Summary\n", "==========================================================================================\n", "Total Bets: 432\n", "Correct B365: 0.33796296296296297\n", "Correct Kalshi: 0.24074074074074073\n", "Flat B365 Returns: $-34.04\n", "Flat Kalshi Returns: $38.02\n", "Flat Kalshi w/ B365 Basis Return: $-22.71\n", "Final B365 Return (Compounded): $1,121.68\n", "Final Kalshi Return (Compounded): $1,541.54\n", "Final Kalshi w/ B365 Basis Return (Compounded): $801.84\n" ] } ], "source": [ "print('Summary')\n", "print('==========================================================================================')\n", "print(f'Total Bets: {binned_df.shape[0]}')\n", "print(f'Correct B365: {binned_df[\"correct_b365\"].mean()}')\n", "print(f'Correct Kalshi: {binned_df[\"correct_kalshi\"].mean()}')\n", "print(f'Flat B365 Returns: ${binned_df[\"returns_b365\"].sum():,.2f}')\n", "print(f'Flat Kalshi Returns: ${binned_df[\"returns_kalshi\"].sum():,.2f}')\n", "print(f'Flat Kalshi w/ B365 Basis Return: ${binned_df[\"returns_kalshi_b365\"].sum():,.2f}')\n", "print(f'Final B365 Return (Compounded): ${binned_df[\"kelly_b365_comp\"].iloc[-1]:,.2f}')\n", "print(f'Final Kalshi Return (Compounded): ${binned_df[\"kelly_kalshi_comp\"].iloc[-1]:,.2f}')\n", "print(f'Final Kalshi w/ B365 Basis Return (Compounded): ${binned_df[\"kelly_kalshi_b365_comp\"].iloc[-1]:,.2f}')\n" ] }, { "cell_type": "code", "execution_count": 30, "id": "64e1e428-5d4b-4c51-96ee-a5595a096bbe", "metadata": {}, "outputs": [], "source": [ "import plotly.graph_objects as go\n", "import numpy as np\n", "\n", "x = np.arange(binned_df.shape[0])\n", "y_0 = binned_df['kelly_b365_comp']\n", "y_1 = binned_df['kelly_kalshi_comp']\n", "y_2 = binned_df['kelly_kalshi_b365_comp']\n", "\n", "fig = go.Figure()\n", "fig.add_trace(go.Scatter(x=x, y=y_0, name='B365', line=dict(color='blue')))\n", "fig.add_trace(go.Scatter(x=x, y=y_1, name='Kalshi', line=dict(color='red')))\n", "fig.add_trace(go.Scatter(x=x, y=y_2, name='Kalshi w/ B365', line=dict(color='green')))\n", "fig.add_trace(go.Scatter(x=bet_df.index, y=bet_df['kelly_kalshi_comp'],\n", " name='Kalshi (ROI-gated, skip)',\n", " line=dict(color='orange', dash='dash'))) # gated overlay\n", "\n", "fig.update_layout(\n", " title='Compounded Returns',\n", " xaxis_title='Bets',\n", " yaxis_title='Returns',\n", " hovermode='x unified'\n", ")\n", "fig.write_html('test_1.html')" ] }, { "cell_type": "code", "execution_count": 31, "id": "3bb2d03a-97fd-419e-b107-47dd85325793", "metadata": {}, "outputs": [], "source": [ "binned_df.to_csv('test.csv', index=False)" ] }, { "cell_type": "markdown", "id": "cbb0f489-e388-497b-bcad-21c3b4469b9d", "metadata": {}, "source": [ "# Early Cashout Testing" ] }, { "cell_type": "code", "execution_count": 32, "id": "10996ac6-9621-4c7b-8ae8-abcf2775f457", "metadata": {}, "outputs": [], "source": [ "df_early = df.dropna(subset=['elapsed'])\n", "\n", "# Pre-compute sell conditions for all (home, away, outcome) combinations\n", "bet_dict = {'H': 'home', 'D': 'draw', 'A': 'away'}\n", "sell_conditions = {}\n", "\n", "for (home, away), group in df_early.groupby(['home', 'away']):\n", " for bet, suffix in bet_dict.items():\n", " key = (home, away, bet)\n", " sell_conditions[key] = int(group[f'kalshi_{suffix}_bid'].max() > 0.8)\n", "\n", "def find_sell_point(row):\n", " \"\"\"If correct, always sell. Otherwise check if bid ever exceeded 0.8.\"\"\"\n", " if row['correct_b365'] == 1:\n", " return 1\n", " return sell_conditions.get((row['home'], row['away'], row['b365_highest_clv']), 0)\n", "\n", "# Apply to DataFrame\n", "df_early['early_sell'] = df_early.apply(find_sell_point, axis=1)" ] }, { "cell_type": "code", "execution_count": 33, "id": "57deb75e-c4e4-42ff-aaf6-bf75e10d4865", "metadata": { "scrolled": true }, "outputs": [], "source": [ "def assign_bin(number):\n", " bins = np.histogram(df_early['elapsed'], bins=num_bins)[1]\n", " for i in range(len(bins)+1):\n", " if number >= bins[i] and number <= bins[i+1]:\n", " return i\n", "df_early = df_early[(df_early['elapsed']>-492.95)&(df_early['elapsed']<=-427.1)]" ] }, { "cell_type": "code", "execution_count": 34, "id": "50ca6252-c75f-4993-b148-eb3f055afcc3", "metadata": {}, "outputs": [], "source": [ "df_early.to_csv('test.csv')" ] }, { "cell_type": "code", "execution_count": null, "id": "23d06e86-1bf0-4247-b74e-0674b48c2417", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "c0167f9d-d6dd-4c53-9899-c3a2d385cd9f", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "7aa69b8b-0e76-478b-92a4-7c552be97daa", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "47e0719d-31b2-40a6-9945-22f19fb1615e", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "id": "significance-testing-header", "metadata": {}, "source": [ "## Significance Testing (Profit-focused on binned_df)" ] }, { "cell_type": "code", "execution_count": 35, "id": "import-stats", "metadata": {}, "outputs": [], "source": [ "from scipy import stats\n", "import numpy as np" ] }, { "cell_type": "code", "execution_count": 36, "id": "onesample-ztest", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "=== One-sample z-tests (Returns > 0?) ===\n", "B365: mean=-0.0788, z=-1.0554, p=0.8544, n=432, Sig: no\n", "Kalshi: mean=0.0880, z=0.8474, p=0.1984, n=432, Sig: no\n", "K+B365: mean=-0.0526, z=-0.6824, p=0.7525, n=432, Sig: no\n", "Kalshi-gated: mean=-0.1308, z=-0.8351, p=0.7982, n=133, Sig: no\n" ] } ], "source": [ "b365_returns = binned_df['returns_b365'].dropna()\n", "kalshi_returns = binned_df['returns_kalshi'].dropna()\n", "kalshi_b365_returns = binned_df['returns_kalshi_b365'].dropna()\n", "gated_returns = bet_df['returns_kalshi'].dropna() # ROI-gated (skip) Kalshi\n", "\n", "def ztest_1samp(data, mu=0, alternative='greater'):\n", " n = len(data)\n", " x_bar = data.mean()\n", " s = data.std(ddof=1)\n", " z = (x_bar - mu) / (s / np.sqrt(n))\n", " if alternative == 'greater':\n", " p = 1 - stats.norm.cdf(z)\n", " elif alternative == 'less':\n", " p = stats.norm.cdf(z)\n", " else:\n", " p = 2 * (1 - stats.norm.cdf(abs(z)))\n", " return z, p\n", "\n", "z_b365, p_b365 = ztest_1samp(b365_returns, alternative='greater')\n", "z_kalshi, p_kalshi = ztest_1samp(kalshi_returns, alternative='greater')\n", "z_kb, p_kb = ztest_1samp(kalshi_b365_returns, alternative='greater')\n", "z_g, p_g = ztest_1samp(gated_returns, alternative='greater')\n", "\n", "print('=== One-sample z-tests (Returns > 0?) ===')\n", "print(f'B365: mean={b365_returns.mean():.4f}, z={z_b365:.4f}, p={p_b365:.4f}, n={len(b365_returns)}, Sig: {\"YES\" if p_b365 < 0.05 else \"no\"}')\n", "print(f'Kalshi: mean={kalshi_returns.mean():.4f}, z={z_kalshi:.4f}, p={p_kalshi:.4f}, n={len(kalshi_returns)}, Sig: {\"YES\" if p_kalshi < 0.05 else \"no\"}')\n", "print(f'K+B365: mean={kalshi_b365_returns.mean():.4f}, z={z_kb:.4f}, p={p_kb:.4f}, n={len(kalshi_b365_returns)}, Sig: {\"YES\" if p_kb < 0.05 else \"no\"}')\n", "print(f'Kalshi-gated: mean={gated_returns.mean():.4f}, z={z_g:.4f}, p={p_g:.4f}, n={len(gated_returns)}, Sig: {\"YES\" if p_g < 0.05 else \"no\"}')" ] }, { "cell_type": "code", "execution_count": 37, "id": "confidence-intervals", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "=== 95% CI for Mean Flat Returns ===\n", "B365: (np.float64(-0.22553789834794646), np.float64(0.06794530575535386))\n", "Kalshi: (np.float64(-0.11612862353237784), np.float64(0.2921453368165689))\n", "K+B365: (np.float64(-0.2039559063208986), np.float64(0.09882684274685932))\n", "Kalshi-gated: (np.float64(-0.4404885902950574), np.float64(0.17896139948512413))\n", "=== Sharpe (per-bet | annualized) ===\n", "B365 per-bet=-0.0508 annual=-3.1002 (3727 bets/yr)\n", "Kalshi per-bet=0.0408 annual=2.4890 (3727 bets/yr)\n", "K+B365 per-bet=-0.0328 annual=-2.0046 (3727 bets/yr)\n", "Kalshi-gated per-bet=-0.0724 annual=-2.4531 (1148 bets/yr)\n", "=== Final Compounded Returns ===\n", "B365: $1,121.68\n", "Kalshi: $1,541.54\n", "K+B365: $801.84\n", "Kalshi-gated: $773.17\n" ] } ], "source": [ "def ci(data):\n", " n = len(data)\n", " mean = data.mean()\n", " std_err = stats.sem(data)\n", " interval = std_err * stats.t.ppf(0.975, n - 1)\n", " return mean - interval, mean + interval\n", "\n", "gated_returns = bet_df['returns_kalshi'].dropna()\n", "\n", "print('=== 95% CI for Mean Flat Returns ===')\n", "print(f'B365: {ci(b365_returns)}')\n", "print(f'Kalshi: {ci(kalshi_returns)}')\n", "print(f'K+B365: {ci(kalshi_b365_returns)}')\n", "print(f'Kalshi-gated: {ci(gated_returns)}')\n", "\n", "# Sharpe \u2014 per-bet, plus annualized by ACTUAL bet frequency (not sqrt(365))\n", "span_days = (binned_df['timestamp'].max() - binned_df['timestamp'].min()) / 86400\n", "bets_per_year = len(binned_df) / span_days * 365\n", "gated_bpy = len(bet_df) / span_days * 365 # gated places fewer bets over the same window\n", "\n", "def sharpe(returns, periods_per_year):\n", " s = returns.std(ddof=1)\n", " if s == 0:\n", " return 0.0, 0.0\n", " per_bet = returns.mean() / s\n", " return per_bet, per_bet * np.sqrt(periods_per_year)\n", "\n", "print('=== Sharpe (per-bet | annualized) ===')\n", "for name, r, bpy in [('B365', b365_returns, bets_per_year),\n", " ('Kalshi', kalshi_returns, bets_per_year),\n", " ('K+B365', kalshi_b365_returns, bets_per_year),\n", " ('Kalshi-gated', gated_returns, gated_bpy)]:\n", " pb, ann = sharpe(r, bpy)\n", " print(f'{name:13} per-bet={pb:.4f} annual={ann:.4f} ({bpy:.0f} bets/yr)')\n", "\n", "print('=== Final Compounded Returns ===')\n", "print(f'B365: ${binned_df[\"kelly_b365_comp\"].iloc[-1]:,.2f}')\n", "print(f'Kalshi: ${binned_df[\"kelly_kalshi_comp\"].iloc[-1]:,.2f}')\n", "print(f'K+B365: ${binned_df[\"kelly_kalshi_b365_comp\"].iloc[-1]:,.2f}')\n", "print(f'Kalshi-gated: ${bet_df[\"kelly_kalshi_comp\"].iloc[-1]:,.2f}')" ] }, { "cell_type": "code", "execution_count": null, "id": "173efa52-3561-4cfd-959c-9f015a5fff24", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 38, "id": "ab756171-b89f-4f83-ad64-4bb65fef3f03", "metadata": {}, "outputs": [], "source": [ "df_filtered_nan_ = df_filtered_nan_[df_filtered_nan_['elapsed']>-90]" ] }, { "cell_type": "code", "execution_count": 39, "id": "d8e0cbc1-3553-4269-8c84-63fb3e698a85", "metadata": {}, "outputs": [], "source": [ "df_last = df_filtered_nan_.sort_values(by='timestamp', ascending=True).drop_duplicates(['home', 'away'], keep='first')" ] }, { "cell_type": "code", "execution_count": 40, "id": "4ddc2498-5717-481a-b01b-340eaed02878", "metadata": {}, "outputs": [], "source": [ "df_last = df_last[(df_last['kalshi_highest_clv_val']>=0.04)&(df_last['kalshi_highest_clv_val']<0.11)]" ] }, { "cell_type": "code", "execution_count": 41, "id": "6e7568a9-e44e-4039-a242-f6a160db997c", "metadata": {}, "outputs": [], "source": [ "# df_last = df_last\n", "df_last['returns'] = df_last.apply(lambda row: row['correct_kalshi']*(1/row['kalshi_'+bet_dict[row['kalshi_highest_clv']]+'_ask'])-1, axis=1)" ] }, { "cell_type": "code", "execution_count": 42, "id": "c2d1eeef-3421-4488-827a-8946327200a0", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "Index(['timestamp', 'game_date', 'league', 'home', 'away', 'status', 'elapsed',\n", " 'home_goals', 'away_goals', 'pin_home', 'pin_draw', 'pin_away',\n", " 'b365_home', 'b365_draw', 'b365_away', 'kalshi_home_mid',\n", " 'kalshi_home_bid', 'kalshi_home_ask', 'kalshi_draw_mid',\n", " 'kalshi_draw_bid', 'kalshi_draw_ask', 'kalshi_away_mid',\n", " 'kalshi_away_bid', 'kalshi_away_ask', 'pin_fair_home', 'pin_fair_draw',\n", " 'pin_fair_away', 'clv_b365_home', 'clv_b365_draw', 'clv_b365_away',\n", " 'clv_kalshi_home', 'clv_kalshi_draw', 'clv_kalshi_away', 'ftr',\n", " 'result_home', 'result_away', 'b365_highest_clv',\n", " 'b365_highest_clv_val', 'kalshi_highest_clv', 'kalshi_highest_clv_val',\n", " 'correct_b365', 'correct_kalshi', 'bin', 'returns'],\n", " dtype='object')" ] }, "execution_count": 42, "metadata": {}, "output_type": "execute_result" } ], "source": [ "df_last.to_csv('test.csv')\n", "df_last.columns" ] }, { "cell_type": "code", "execution_count": null, "id": "5b1cfa54-eea6-461b-b25f-4322351a6e25", "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": "3.10.11" } }, "nbformat": 4, "nbformat_minor": 5 }