BACKTEST RESULTS — MARCH 2026

BTC Dropped 20%. AEGIS Made Money on ETH.

We backtested the AEGIS signal engine on 365 days of real market data across 10 cryptocurrencies. No cherry-picking, no hypothetical trades. Every entry, every exit, every loss — all documented below.

The Market We Tested In

This wasn't a friendly market. Between March 2025 and March 2026:

-20%
Bitcoin
-48%
Dogecoin
-35%
Solana
-64%
Cardano

Most crypto traders lost money. Most trading bots lost money. The average retail crypto portfolio dropped 30-50% during this period.

AEGIS didn't just survive — it turned a profit on ETH while the rest of the market bled.

The Results

Configuration tested: 4-hour candles, 1.0x ATR stop-loss, 8.0x ATR take-profit, $3,000 starting balance, 10% position size per trade. No time-based exits. Pure signal-driven entries with tight stops and wide targets.

ETH-USD: The Winner

82
Total Trades
16%
Win Rate
+14.6%
Avg Win
-2.3%
Avg Loss
1.20
Profit Factor
$3,092
Final Balance

16% win rate sounds bad. But the wins are 6x larger than the losses. Lose $2.30 on a bad trade. Make $14.60 on a good one. Over 82 trades, that adds up.

This is the "lose small, win big" approach used by professional trend followers. Most trades lose — but the winners more than make up for it.

Why ETH and Not Everything Else?

CoinPrice ChangeAEGIS ResultProfit FactorVerdict
ETH-USD+7.0%+3.1%1.20Profitable
LINK-USD-39.1%-2.0%0.87Near breakeven
AVAX-USD-55.8%-2.7%0.88Near breakeven
BTC-USD-20.4%-3.6%0.70Small loss
XRP-USD-36.7%-3.3%0.79Small loss
SOL-USD-34.8%-7.9%0.59Loss

ETH was the only coin with positive price movement over the period. The signal engine performs best on assets that have some upward momentum — it's not magic, it's math. In a bear market, the bot limits losses. In a bull market, it captures gains.

Notice something: even on coins that dropped 35-56%, AEGIS only lost 2-8%. Compare that to holding — you would have lost 35-64%. The tight stop-loss (1.0x ATR) cuts losers fast and preserves capital.

The Fee Problem

The single biggest factor in whether a trading bot makes or loses money isn't the signals — it's the fees.

ExchangeFee (round-trip)ETH ResultProfitable?
Gemini (maker)0.4%+3.1%Yes
Gemini (taker)0.8%+1.5%Yes
Coinbase (maker)0.8%+1.5%Yes
Coinbase (taker)1.2%-7.4%No
Robinhood1.7%-14.1%No

The same signals, the same entries, the same exits. The only difference is what the exchange charges. At Gemini's maker rate (0.2% per trade), AEGIS makes money. At Robinhood's 0.85% spread, it loses money.

We're being honest about this. Most trading bots hide their fee assumptions or show backtests without fees. We ran every test with real fees baked into every trade — entry and exit. If it's not profitable after fees, we say so.

How the Signal Engine Works

AEGIS uses 7 technical indicators scored on a weighted system:

  1. RSI (Relative Strength Index) — identifies oversold and overbought conditions
  2. MACD — detects momentum shifts and trend changes
  3. EMA 20/50 — confirms short and medium-term trend direction
  4. Bollinger Bands — measures volatility and price extremes
  5. Volume — confirms whether moves have conviction behind them
  6. Support/Resistance — identifies key price levels from historical swing points
  7. Z-Score — detects sharp price drops that may snap back

Each indicator contributes points to a confidence score. A BUY signal fires when bullish points significantly outweigh bearish points. An AI validation layer then checks market context (sentiment, funding rates, BTC trend) and can reject signals that look technically good but structurally bad.

Directional Accuracy

Ignoring fees entirely, how often does the signal engine correctly predict that price will go up?

CoinSignal AccuracyRandom Would BeEdge
BTC-USD50.9%47.1%+3.8%
ETH-USD49.6%47.1%+2.5%
DOGE-USD48.5%47.1%+1.4%
XRP-USD44.5%47.1%-2.6%
SOL-USD46.1%47.1%-1.0%

A 2-4% edge over random doesn't sound like much. But combined with asymmetric risk/reward (losing 2.3% when wrong, gaining 14.6% when right), it compounds into real profit. This is how professional quant funds operate — small edge, repeated hundreds of times.

What We Learned

1. Fees are everything

The difference between a profitable bot and a losing bot is often just the exchange fee. We originally built AEGIS for Robinhood and spent months wondering why it kept losing. The signals were right 51% of the time — but 0.85% per trade ate every win.

2. Lose small, win big

A 16% win rate sounds terrible. But when your average win is 6x your average loss, you only need to be right 1 out of 7 times to break even. This strategy wins on the size of wins, not the number of wins.

3. Not every coin works

The signal engine is profitable on ETH, near-breakeven on LINK and AVAX, and unprofitable on SOL and DOGE. We don't pretend otherwise. Different coins have different characteristics, and not all respond to technical analysis the same way.

4. Bear markets are survivable

While buy-and-hold lost 20-64% depending on the coin, AEGIS limited losses to 2-8% on losing coins and made money on ETH. The tight stop-loss is the key — it cuts losers before they become disasters.

Methodology

The AI Validator: Honest Results

We built an AI layer that reviews every trade signal before execution. The idea: catch structurally dangerous setups that the signal engine misses. We backtested it on all 355 ETH signals from the past year. Here's what happened.

Full-Year AI Backtest (355 Trades)

101
Trades Rejected
84%
Rejection Accuracy
-90.8%
PnL With AI
-58.1%
PnL Without AI

The AI made things worse. We're publishing this because honesty matters more than marketing.

The AI correctly blocked 85 out of 101 rejected trades (84% accuracy). Those 85 trades would have lost money. On paper, that sounds great.

But it also blocked 16 winning trades — including several of the big +12% to +17% winners that the "lose small, win big" strategy depends on. In a system where only 14% of trades win, blocking even a handful of winners is devastating. The math doesn't recover.

Why the AI Hurts This Strategy

The "lose small, win big" strategy works because rare, large winners offset many small losses. The AI sees a risky-looking setup and blocks it. But risky-looking setups ARE the ones that produce the big wins — they're momentum trades in volatile conditions. Blocking them removes the engine's edge.

This is a known problem in quantitative trading called the overfitting paradox: a filter that removes losing trades also removes the winners that share similar characteristics. The AI can't distinguish between a -3% loser and a +15% winner at entry time — they look the same going in.

What We're Doing About It

What the AI Does Well

The Full Picture

After a full year of backtesting, here's what actually drives performance:

LayerWhat It DoesImpact
Signal EnginePredicts direction 51% of the timeSmall but real edge over random
Asymmetric ExitsTight stop (1x ATR), wide target (8x ATR)Wins are 6x larger than losses
Low Fees (Gemini)0.2% per trade vs Robinhood's 0.85%Difference between profit and loss
AI ValidatorReviews every trade with reasoningBetter as a monitoring tool than a hard filter

Three layers make the system profitable: accurate signals, asymmetric risk/reward, and low fees. The AI is a work in progress — it adds transparency and catches extreme risks, but shouldn't hard-block trades in a strategy designed for rare big wins.

Risk Warning: Past performance does not guarantee future results. Cryptocurrency trading carries significant risk. These backtests simulate historical performance and may not reflect live trading conditions including slippage, liquidity gaps, and market impact. The AI validator backtest covered 355 trades over 365 days — a meaningful sample size, and the results show it hurts more than it helps as a hard filter. We're being transparent about this. Live results will vary. Never trade with money you cannot afford to lose.

See It for Yourself

Start paper trading with a $10K virtual bankroll. Watch the signals. Watch the AI reject bad setups. Go live only when you're convinced.

START PAPER TRADING — FREE