Trend-follow — EMA crossover¶

The simplest momentum strategy: buy when a fast moving average crosses above a slower one, exit on the reverse cross. With a fixed bracket so a bad trade doesn't run.

What it does¶

Trades Bitcoin on 1-minute candles
Buys when EMA(9) crosses above EMA(21) — momentum has shifted up
Sells when EMA(9) crosses below EMA(21) — momentum has shifted down
Bracket caps each trade at a 50% loss / 100% gain (relative to the position's stop distance, not the position value)
One position at a time — won't pyramid into the trade

This is the textbook trend-follower. It catches sustained moves; it underperforms in choppy markets where the EMAs whipsaw.

The strategy file¶

strategies/trend-follow.qkt

STRATEGY trend_follow VERSION 1

LET fast = 9
LET slow = 21

SYMBOLS
    btc = BACKTEST:BTCUSDT EVERY 1m

RULES
    WHEN ema(btc.close, fast) CROSSES ABOVE ema(btc.close, slow)
    THEN BUY btc SIZING 0.1
         BRACKET {
           STOP_LOSS BY 50 PCT,
           TAKE_PROFIT BY 100 PCT
         }
         LOG "long entry" fast=ema(btc.close, fast) slow=ema(btc.close, slow)

    WHEN ema(btc.close, fast) CROSSES BELOW ema(btc.close, slow)
    THEN CLOSE btc
         LOG "exit on cross-below"

How to run it¶

# Fetch a month of Bitcoin data if you don't have it
./scripts/fetch-dukascopy.sh BTCUSDT 2024-01-01 2024-02-01

# Backtest
qkt backtest strategies/trend-follow.qkt --from 2024-01-01 --to 2024-02-01

# Paper-trade
qkt run strategies/trend-follow.qkt

What to expect¶

A typical month of BTC at 1-minute resolution produces 8–15 trades with the default (9, 21) parameters. Win rate hovers around 40–50% — trend-following strategies make money by letting winners run, not by winning often.

Trades:           14
Final realized:   234.50
Win rate:         0.571
Sharpe (daily):   1.34
Max drawdown:     -180.25

If your numbers are wildly different, check the date range and that the data populated correctly (./scripts/fetch-dukascopy.sh BTCUSDT ... writes to ~/.qkt/data/symbols/BTCUSDT/).

How to adapt it¶

Different fast/slow periods¶

Edit the LET lines:

LET fast = 12
LET slow = 26

Or sweep them — see Run a parameter sweep.

Different timeframe¶

Change EVERY 1m to EVERY 5m, EVERY 15m, EVERY 1h, etc. 1m generates noise; 1h smooths it. Most traders find 15m–1h is the sweet spot for crypto trend-follow.

Different asset¶

Swap the symbol declaration:

SYMBOLS
    eur = BACKTEST:EURUSD EVERY 15m

Then reference eur instead of btc in the rules. The strategy logic is symbol-agnostic.

Different bracket¶

The 50 PCT / 100 PCT is percent-of-entry-price, not percent-of-position. To set absolute prices instead:

BRACKET {
  STOP_LOSS AT btc.close * 0.95,    -- 5% below entry
  TAKE_PROFIT AT btc.close * 1.10   -- 10% above entry
}

Add a volatility filter¶

Trend-followers do badly in choppy ranges. Gate entries by ATR:

LET vol = atr(btc, 14)

RULES
    WHEN ema(btc.close, fast) CROSSES ABOVE ema(btc.close, slow)
     AND vol > 50  -- only enter when there's enough movement
    THEN BUY btc SIZING 0.1 BRACKET { STOP_LOSS BY 50 PCT, TAKE_PROFIT BY 100 PCT }

Common gotchas¶

First-bar bug. If your start date happens to be already in an uptrend (EMA9 > EMA21 on bar 1), the rule won't fire until the next cross. This is correct edge-trigger behaviour, but surprises people. To trade from the start, add a different entry rule or backfill a few days of warmup.
EMAs need warmup. The first ~21 bars produce null EMA values; rules using them won't fire. The compiler infers warmup automatically from the 21 in ema(..., 21). (An explicit WARMUP N BARS declaration is on the roadmap.)
Bracket fills aren't free. Even though they appear atomic in the DSL, the broker may fill the stop with slippage on a fast move. Paper-test for at least a few weeks before live deployment.

What this example demonstrates¶

Single stream declaration (btc = BACKTEST:BTCUSDT EVERY 1m)
Indicator function calls (ema(btc.close, 9))
Edge-triggered conditions (CROSSES ABOVE)
Atomic bracket orders
LOG actions for inline debugging
LET variables for tunable parameters

See the DSL reference for everything else.