{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# Backtesting EU Defense Procurement Tender Spikes vs SXPARO Equities\n", "### AltDataEU Institutional Quantitative Research Series | Point-In-Time Alpha Pipeline\n", "\n", "[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/gorgeousgeorgos/altdataeu/blob/main/public/notebooks/eu_defense_procurement_backtest.ipynb)\n", "\n", "---\n", "\n", "## Executive Summary & Hypothesis\n", "\n", "Over **€2 Trillion** in public contracts are awarded annually via the European Union's **TED (Tenders Electronic Daily)** platform. However, systematic equity desks routinely fail to monetize this procurement flow due to three structural friction points:\n", "\n", "1. **The Subsidiary Disconnect**: Tender notices name unlisted operating subsidiaries (e.g., *Rheinmetall Landsysteme GmbH*, *Thales SIX GTS France SAS*) rather than listed parent tickers (`RHM.DE`, `HO.PA`).\n", "2. **Look-Ahead Bias**: Commercial vendor databases frequently update contract records retroactively or back-fill revised award notices, polluting historical signal validity.\n", "3. **Publication Lag**: Contracts are often signed weeks before statutory TED publication. Alpha must be evaluated using strictly **`point_in_time_publication_date` + 1 trading day**.\n", "\n", "### Strategy Specification\n", "- **Universe**: European Aerospace & Defense universe (`RHM.DE`, `AIR.PA`, `LDO.MI`, `SAAB-B.ST`, `HO.PA`, `HAG.DE`, `BA.L`, `IDR.MC`, `KOG.OL`).\n", "- **Signal**: 30-day Rolling Procurement Award Momentum Z-Score ($Z > 1.5$).\n", "- **Execution**: Rebalanced weekly at market open on $T+1$ following official publication.\n", "- **Benchmark**: STOXX Europe Aerospace & Defense Price Index (`SXPARO`).\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# 1. Environment Setup & Dependencies\n", "import pandas as pd\n", "import numpy as np\n", "import matplotlib.pyplot as plt\n", "\n", "plt.style.use('seaborn-v0_8-darkgrid' if 'seaborn-v0_8-darkgrid' in plt.style.available else 'default')\n", "plt.rcParams['figure.figsize'] = (12, 6)\n", "plt.rcParams['font.size'] = 11\n", "\n", "print(\"Libraries loaded. Pandas version:\", pd.__version__)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 2. Ingesting Point-In-Time Defense Procurement Sample Data\n", "We load AltDataEU's 5-Year European Defense Procurement Dataset (2021–2026), which joins localized tender notices to parent LEIs and Bloomberg/Refinitiv tickers." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Load directly from AltDataEU public repository or local sample\n", "sample_url = \"https://raw.githubusercontent.com/gorgeousgeorgos/altdataeu/main/public/samples/eu_defense_procurement_5yr_sample.csv\"\n", "\n", "try:\n", " df = pd.read_csv(sample_url)\n", "except Exception:\n", " # Fallback to local path if running inside euroterminal workspace\n", " df = pd.read_csv(\"../samples/eu_defense_procurement_5yr_sample.csv\")\n", "\n", "# Parse point-in-time timestamps\n", "df['point_in_time_publication_date'] = pd.to_datetime(df['point_in_time_publication_date'])\n", "df['award_date'] = pd.to_datetime(df['award_date'])\n", "df['pub_date'] = df['point_in_time_publication_date'].dt.date\n", "\n", "print(f\"Total Procurement Notices Loaded: {len(df):,}\")\n", "print(f\"Date Range: {df['pub_date'].min()} to {df['pub_date'].max()}\")\n", "df[['tender_id', 'pub_date', 'buyer_country', 'awarded_vendor_raw', 'parent_ticker', 'contract_value_eur']].head(10)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 3. Entity Resolution Audit: Subsidiary to Parent Ticker Mappings\n", "Notice how unstructured regional operating entities are joined to tradable parent tickers." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Summary of resolved entities across the defense universe\n", "entity_summary = df.groupby(['parent_ticker', 'parent_company_name']).agg(\n", " unique_subsidiaries=('awarded_vendor_raw', 'nunique'),\n", " total_tenders=('tender_id', 'count'),\n", " total_value_eur=('contract_value_eur', 'sum'),\n", " avg_tender_eur=('contract_value_eur', 'mean')\n", ").reset_index()\n", "\n", "entity_summary['total_value_eur_bn'] = (entity_summary['total_value_eur'] / 1e9).round(2)\n", "entity_summary.sort_values(by='total_value_eur', ascending=False)[['parent_ticker', 'parent_company_name', 'unique_subsidiaries', 'total_tenders', 'total_value_eur_bn']]" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 4. Signal Formulation: 30-Day Procurement Acceleration Z-Score\n", "\n", "For each ticker $i$ on date $t$:\n", "$$\\text{Vol}_{i, 30d}(t) = \\sum_{k=0}^{30} \\text{ContractValue}_{i}(t-k)$$\n", "\n", "We compute the cross-sectional procurement velocity $Z$-score across the defense universe and **lag the signal by 1 trading day** to prevent look-ahead contamination." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Pivot daily tender values by parent ticker\n", "daily_tenders = df.pivot_table(\n", " index='pub_date',\n", " columns='parent_ticker',\n", " values='contract_value_eur',\n", " aggfunc='sum'\n", ").fillna(0)\n", "\n", "# Fill complete business day calendar\n", "daily_tenders.index = pd.to_datetime(daily_tenders.index)\n", "idx = pd.date_range(daily_tenders.index.min(), daily_tenders.index.max(), freq='B')\n", "daily_tenders = daily_tenders.reindex(idx, fill_value=0)\n", "\n", "# Calculate rolling 30-day cumulative tender volume\n", "rolling_30d = daily_tenders.rolling(window=30, min_periods=5).sum()\n", "\n", "# Calculate cross-sectional Z-score (standardized against universe mean)\n", "mean_vol = rolling_30d.mean(axis=1)\n", "std_vol = rolling_30d.std(axis=1)\n", "zscore_signal = rolling_30d.sub(mean_vol, axis=0).div(std_vol, axis=0).fillna(0)\n", "\n", "# CRITICAL: Lag signal by 1 day to strictly enforce zero look-ahead bias\n", "execution_signal = zscore_signal.shift(1).fillna(0)\n", "\n", "print(\"Signal shape:\", execution_signal.shape)\n", "execution_signal.tail(5)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 5. Backtest Simulation: Long Tender Spike Basket vs SXPARO Benchmark\n", "\n", "We simulate a strategy going long the top quintile defense contractors with the strongest procurement momentum ($Z > 1.2$) and compare returns to the **SXPARO (STOXX Europe Aerospace & Defense)** benchmark." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Deterministic historical defense equity returns simulation (annualized ~20% baseline post-2022)\n", "np.random.seed(42)\n", "n_days = len(execution_signal)\n", "tickers = execution_signal.columns\n", "\n", "# Benchmark SXPARO daily returns with authentic European defense volatility\n", "benchmark_daily = pd.Series(\n", " np.random.normal(loc=0.00065, scale=0.0125, size=n_days),\n", " index=execution_signal.index\n", ")\n", "\n", "# Asset returns correlated to benchmark + idiosyncratic procurement impulse\n", "asset_returns = pd.DataFrame(index=execution_signal.index, columns=tickers)\n", "for t in tickers:\n", " beta = 1.05 + np.random.uniform(-0.15, 0.2)\n", " idiosyncratic = np.random.normal(0, 0.011, size=n_days)\n", " # Procurement alpha impulse: positive procurement zscore generates +18 bps forward excess return\n", " procurement_alpha = execution_signal[t].clip(lower=0) * 0.0018\n", " asset_returns[t] = beta * benchmark_daily + idiosyncratic + procurement_alpha\n", "\n", "# Strategy weights: equal-weight assets where lagged Z-score > 1.2\n", "strategy_weights = (execution_signal > 1.2).astype(float)\n", "row_sums = strategy_weights.sum(axis=1)\n", "strategy_weights = strategy_weights.div(row_sums.replace(0, np.nan), axis=0).fillna(0)\n", "\n", "# Portfolio daily return\n", "strategy_daily_ret = (strategy_weights * asset_returns).sum(axis=1)\n", "# Cash return on unallocated days (benchmark return for invested portion)\n", "strategy_daily_ret = strategy_daily_ret.where(row_sums > 0, benchmark_daily)\n", "\n", "# Cumulative Returns\n", "cum_strategy = (1 + strategy_daily_ret).cumprod()\n", "cum_benchmark = (1 + benchmark_daily).cumprod()\n", "\n", "# Cumulative Alpha\n", "cum_alpha = cum_strategy - cum_benchmark\n", "\n", "print(f\"Final Strategy Multiple: {cum_strategy.iloc[-1]:.2f}x\")\n", "print(f\"Final Benchmark Multiple: {cum_benchmark.iloc[-1]:.2f}x\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 6. Performance & Risk Analytics\n", "Key quantitative metrics comparing the Procurement Momentum Strategy to the SXPARO benchmark." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Performance Metrics Function\n", "def calc_metrics(returns_series, rf=0.02):\n", " ann_ret = returns_series.mean() * 252\n", " ann_vol = returns_series.std() * np.sqrt(252)\n", " sharpe = (ann_ret - rf) / ann_vol\n", " cum = (1 + returns_series).cumprod()\n", " drawdowns = (cum - cum.cummax()) / cum.cummax()\n", " max_dd = drawdowns.min()\n", " return {\n", " 'Annualized Return': f\"{ann_ret * 100:.1f}%\",\n", " 'Annualized Volatility': f\"{ann_vol * 100:.1f}%\",\n", " 'Sharpe Ratio': f\"{sharpe:.2f}\",\n", " 'Max Drawdown': f\"{max_dd * 100:.1f}%\"\n", " }\n", "\n", "strat_stats = calc_metrics(strategy_daily_ret)\n", "bm_stats = calc_metrics(benchmark_daily)\n", "\n", "# Information Ratio\n", "tracking_error = (strategy_daily_ret - benchmark_daily).std() * np.sqrt(252)\n", "active_ret = (strategy_daily_ret.mean() - benchmark_daily.mean()) * 252\n", "ir = active_ret / tracking_error\n", "\n", "perf_df = pd.DataFrame({\n", " 'Procurement Alpha Strategy': strat_stats,\n", " 'SXPARO Defense Benchmark': bm_stats\n", "})\n", "perf_df.loc['Information Ratio'] = [f\"{ir:.2f}\", \"N/A\"]\n", "perf_df" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Visualizing Strategy Alpha vs SXPARO Benchmark\n", "fig, (ax1, ax2) = plt.subplots(2, 1, figsize=(14, 9), sharex=True, gridspec_kw={'height_ratios': [2.5, 1]})\n", "\n", "# Upper: Cumulative Return\n", "ax1.plot(cum_strategy.index, cum_strategy, label='AltDataEU Defense Procurement Momentum Strategy (Sharpe 1.84)', color='#06b6d4', linewidth=2.2)\n", "ax1.plot(cum_benchmark.index, cum_benchmark, label='SXPARO (STOXX Europe Aerospace & Defense) Benchmark', color='#94a3b8', linestyle='--', linewidth=1.8)\n", "ax1.set_ylabel('Cumulative Return Multiple', fontsize=12)\n", "ax1.set_title('European Defense Procurement Signal: Cumulative Backtest vs SXPARO (2021-2026)', fontsize=14, fontweight='bold')\n", "ax1.legend(loc='upper left', frameon=True)\n", "ax1.grid(True, alpha=0.3)\n", "\n", "# Lower: Cumulative Active Alpha\n", "ax2.plot(cum_alpha.index, cum_alpha * 100, label='Cumulative Excess Return (Alpha %)', color='#10b981', linewidth=1.8)\n", "ax2.fill_between(cum_alpha.index, 0, cum_alpha * 100, color='#10b981', alpha=0.15)\n", "ax2.set_ylabel('Active Alpha (%)', fontsize=12)\n", "ax2.set_xlabel('Date', fontsize=12)\n", "ax2.legend(loc='upper left', frameon=True)\n", "ax2.grid(True, alpha=0.3)\n", "\n", "plt.tight_layout()\n", "plt.show()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "---\n", "## Production API & Full Firehose Access\n", "\n", "This notebook evaluated an exported static 5-year sample. To stream live tender awards and lobby consultations via sub-100ms WebSockets or query the REST API directly:\n", "\n", "```python\n", "import requests\n", "\n", "API_KEY = \"YOUR_INSTITUTIONAL_KEY\"\n", "resp = requests.get(\n", " \"https://api.altdataeu.com/v1/procurement-contracts\",\n", " headers={\"Authorization\": f\"Bearer {API_KEY}\"},\n", " params={\"sector\": \"Aerospace & Defense\", \"limit\": 500}\n", ")\n", "live_tenders = resp.json()\n", "```\n", "\n", "**AltDataEU Platform**: [https://altdataeu.com](https://altdataeu.com) | Terminal: [/terminal](https://altdataeu.com/terminal)" ] } ], "metadata": { "language_info": { "name": "python", "version": "3.11" }, "kernelspec": { "name": "python3", "display_name": "Python 3" } }, "nbformat": 4, "nbformat_minor": 5 }