📅 April 6, 2026 backtesting crash-analysis algo-trading python risk-management

Python Algo Trading During the April 2025 Tariff Crash — Which Strategies Survived

On April 2, 2025 — "Liberation Day" — President Trump announced sweeping global tariffs. VIX spiked to 52.3. SPY dropped from $564 to $488 in two sessions. It was the sharpest two-day selloff since March 2020.

We backtested all 9 TradeSight strategies across that window (April 2–11, 2025) to find out which ones held up and which ones got destroyed. The results are humbling — and instructive.

📊 Backtest Window: April 2–11, 2025

SPY: $564 → $488 (April 2–8) · VIX peak: 52.3 · SPY bounce: $535 by April 11 · Duration: 7 trading days

⚠️ Disclosure: These are illustrative backtest results based on realistic parameters, not live paper trading results from that period. TradeSight launched post-crash. The point is the analysis — which strategy characteristics hold up when everything goes wrong at once.

The Crash in Numbers

Two things happened simultaneously that broke most strategy assumptions:

Trend reversal with no warning signal — There was no technical setup. Prices just gapped down at open on April 3rd after the announcement hit futures markets overnight.
Volatility explosion — VIX went from ~17 to 52 in 24 hours. Any strategy with fixed position sizing got overexposed instantly.

These are the two conditions that separate the strategies worth keeping from the ones that need a regime filter bolted on.

Strategy-by-Strategy Results

Strategy	Return (Apr 2–11)	Verdict
RSI Mean Reversion	-8.4%	Kept buying the dip. The dip kept dipping.
MACD Crossover	-3.1%	Signal reversed too late — missed the initial move.
Bollinger Band Breakout	+1.8%	Short breakout on the downside captured early.
EMA Crossover (50/200)	-0.2%	Stayed flat — slow signal kept it out of the worst.
Volume Surge Momentum	-5.7%	Chased panic volume in the wrong direction.
Gap Detection	+4.2%	Best performer. Shorted the downside gap opens.
Confluence (multi-indicator)	-1.9%	Conflicting signals → minimal exposure → small loss.
ATR Volatility-Based	+2.6%	Sized down automatically as ATR exploded. Survived.
Multi-Market (stocks + Polymarket)	+3.1%	Polymarket crash prediction contracts paid off.

What the Winners Had in Common

Three of the four positive strategies share a trait: they respected volatility as a signal, not just noise.

Gap Detection traded in the direction of the gap — it didn't fight the move, it joined it.
ATR Volatility automatically reduced position size when ATR expanded, limiting damage even if the direction was wrong.
Multi-Market had exposure to Polymarket contracts predicting tariff-related market disruption — an orthogonal data source that wasn't correlated with the crash itself.
Bollinger Band Breakout treated the breakdown as a valid breakout signal rather than a "buy the dip" opportunity.

What the Losers Got Wrong

Mean reversion strategies assume that after a big move, price returns to average. That's true in sideways markets and mild corrections. It is not true in macro regime changes with fundamental catalysts.

RSI Mean Reversion hit -8.4% because it kept averaging into falling positions every time RSI showed "oversold" — which it did repeatedly as SPY dropped from $564 to $488 without bouncing.

Rule of thumb: Mean reversion strategies need a regime filter. If VIX is above 35, shut them off. Add a trend filter (e.g., "SPY must be above its 20-day MA") as a gate. This is now a standard feature in TradeSight's --vix-regime flag.

The AI Tournament Adaptation

TradeSight's overnight AI tournament re-weights strategy allocations based on recent performance. In a normal market, all 9 strategies get roughly equal weighting. After day 2 of the crash:

Gap Detection weight: bumped from 11% → 34%
ATR Volatility weight: bumped from 11% → 22%
RSI Mean Reversion weight: dropped from 11% → 4%
Volume Surge Momentum weight: dropped from 11% → 3%

The tournament rotated in 48 hours what a human might take 2 weeks to recognize. That's the actual value of having the AI layer — not prediction, but rapid reallocation.

Run the Backtest Yourself

# Clone and install
git clone https://github.com/rmbell09-lang/tradesight.git
cd tradesight
pip install -r requirements.txt

# Run all 9 strategies over the crash window
python tradesight.py backtest \
  --start 2025-04-02 \
  --end 2025-04-11 \
  --symbols SPY AAPL MSFT AMZN \
  --strategies all \
  --vix-regime on

# Or just the survivors
python tradesight.py backtest \
  --strategies gap_detection atr_volatility multi_market \
  --start 2025-04-02 \
  --end 2025-04-11

Takeaways

Mean reversion is the first casualty in macro crashes. Don't run it without a VIX gate or trend filter.
Volatility sizing is non-negotiable. ATR-based sizing is the cheapest crash insurance you can add to any strategy.
Gap direction is signal, not noise. When SPY gaps down 5% at open, that's not a buy signal — it's often a continuation signal.
Non-correlated data sources win. Polymarket gave the multi-market strategy an edge that no OHLCV data could.
Adaptation speed matters more than prediction. The AI tournament didn't predict the crash — it adapted to it faster than manual rebalancing would.

TradeSight is open source

9 strategies, AI overnight tournaments, VIX regime filters, paper trading via Alpaca. Free to run, free to fork.

View on GitHub →