Strategies and backtesting

A strategy is an indicator that also says when to be in the market. Swap indicator( for strategy(, return entry and exit signals alongside your plots, and the workbench replays them into an equity curve, a trade ledger, and a summary. Then the same script can be deployed as a bot.

The smallest strategy

from pairlens import strategy, series, marker
from pairlens.ta import ema, crossover, crossunder

meta = strategy(
    title='EMA Cross',
    series=[series.line('fast'), series.line('slow')],
    markers=[marker.buy('enter_long'), marker.sell('enter_short')],
    initial_capital=10000.0,
    position_size=1.0,
    fee=0.001,
    stop_loss=0.03,
    take_profit=0.06,
    min_bars=60,
)


def compute(ctx):
    fast, slow = ema(ctx.close, 21), ema(ctx.close, 55)
    return {
        'fast': fast,
        'slow': slow,
        'enter_long': crossover(fast, slow),
        'enter_short': crossunder(fast, slow),
        'position': ...,  # see below
    }

Signalling position

You can express intent four ways, and the backtester reads whichever you provide:

position is the clearest when your strategy always knows what it wants to hold. entries and exits are the clearest when your strategy thinks in events. Pick one style and stay in it.

Any array that is missing or shorter than the window reads as zero, so a script that only fills part of the window degrades to “no signal” instead of throwing.

Protective exits

Declare them in meta, not in compute():

stop_loss=0.03,       # 3% below entry
take_profit=0.06,     # 6% above entry
trailing_stop=0.02,   # 2% off the best price reached
max_bars=48,          # force out after 48 bars

All four are fractions of the entry price. They are evaluated per bar against the held position by @pairlens/bot-engine/risk, the exact module a live bot runs. The backtester does not implement its own stop logic. That is deliberate: a tester with its own stop would eventually disagree with the bot it is meant to predict, and the disagreement would first show up on the bar where real money was at stake.

Write a strategy that has never seen stop_loss= and you will meet stops for the first time with real money on the table. Declare them early.

The rule that makes the numbers real

A signal on bar i fills at the open of bar i + 1.

Filling at the signal bar’s close would let the strategy trade on information it could not have had, which turns every backtest into fiction. A consequence: a signal on the very last bar never fills.

Fees and slippage are charged on both legs, in the direction that hurts you.

Reading the report

Run the script and the Backtest panel appears next to the preview chart.

Beating buy and hold is the bar that matters in a bull market. A strategy with a 70% win rate and a profit factor of 0.9 is losing money slowly, and the win rate is why it feels good while doing it.

The trade log

Every round trip is listed with direction, entry, exit, size, P&L, and an exit reason:

signal (the strategy changed its mind), stop-loss, take-profit, trailing-stop, max-bars, or open for the position that ran out of data rather than out of reasons.

A live bot’s event log uses the same words for the same events, so a tester row and a bot row read identically.

How to not fool yourself

Vary the window. Change the history depth and the pair. A strategy that only works on 2,000 bars of one symbol found a pattern in that symbol, not in markets.

Watch the trade count. Twelve trades is a story, not a sample.

Change one parameter at a time. The parameter panel makes sweeping easy, which also makes overfitting easy. If the equity curve collapses when a length goes from 20 to 21, you have fitted noise.

Look at lookahead. If you used align(..., lookahead=True) anywhere, the backtest is reading the future and the number is meaningless.

Then paper trade it. A bot in paper mode running forward on live data is the only test that cannot be overfitted, because the data has not happened yet.

From strategy to bot

Once the numbers hold up, deploy it. The workbench header shows a Deploy as bot button on any strategy script that has run: one click opens the bot create flow with that script preselected. Pick a market and timeframe, set sizing and guards, and it starts on paper. Going live is a separate, explicit gate. The same flow is reachable from Bots → New bot, where your strategies sit next to the ready-made ones. See bots.