No-Code Trading Strategy Backtesting Tutorial

This tutorial builds one no-code backtest from start to finish. You will define a plain-English thesis, inspect the rule Stingray creates, freeze the replay settings, and decide whether the evidence supports rejection, refinement, or live monitoring.

No capital or execution permission is part of the exercise.

The exercise

Test this thesis:

When BTC funding on Hyperliquid is negative while Binance BTCUSDT rises at least 0.5% over 60 minutes, crowded shorts may be under pressure.

Use this complete prompt:

Backtest BTC from 2026-04-20 through 2026-05-18. Match when Hyperliquid funding is below 0 bps per hour and Binance BTCUSDT is up at least 0.5% over the prior 60 minutes. Use five-minute evaluation, a one-hour cooldown, and show 1-hour, 4-hour, and 24-hour forward returns. Keep the rule as a draft and do not activate it.

Step 1: check the typed rule

Before running anything, verify that the generated definition preserves:

Field	Expected value
Market	BTC
Funding source	Hyperliquid
Price source	Binance BTCUSDT
Funding threshold	Below 0 bps/hour
Momentum threshold	At least +0.5% over 60 minutes
Evaluation interval	Five minutes
Cooldown	One hour
Replay window	2026-04-20 to 2026-05-18
Activation	Draft only

If the rule substitutes a venue, changes a threshold, or omits the cooldown, correct it before the replay. A fluent summary is not a substitute for the exact condition.

Step 2: check data and time

Hyperliquid documents that funding payments occur hourly and exposes historical funding through its public information API. Those source semantics matter: the replay should not treat an hourly funding observation as if it were a new five-minute value. See the official Hyperliquid funding documentation and historical funding endpoint.

Confirm the artifact names both sources, discloses missing ranges, and does not use future observations to decide whether an earlier condition fired.

Step 3: separate matches from events

A condition may remain true over several consecutive rows. Read three counts separately:

• Raw matches: every sample where the rule is true.
• Distinct fires: transitions into the full condition.
• After cooldown: fires that would actually produce a notification.

The published version of this exercise found 150 matching five-minute windows and 37 notifications after the one-hour cooldown. That does not mean 150 independent opportunities existed.

Step 4: read forward returns

The published artifact reported a 24-hour average forward return of +0.41%, a +0.55% median, and a 62% positive share across eligible observations. A non-overlapping 24-hour illustration compounded to +8.3% versus +4.9% for BTC buy-and-hold over the same window.

These are historical outcome measurements, not achievable P&L. They exclude a complete model of entries, exits, size, leverage, fees, spread, slippage, and market impact. The sample covers one short window and may represent one regime.

Read the full funding-rate walkthrough and the Backtesting Benchmark Report for the published evidence table.

Step 5: stress the idea

Before monitoring, create genuinely separate tests:

• Move the replay to a later window without changing the rule.
• Compare negative funding alone with funding plus momentum.
• Test whether the result is dominated by a single volatility episode.
• Add realistic fees and slippage outside the signal replay.
• Compare BTC with another liquid market without assuming the behavior transfers.

Do not optimize thresholds repeatedly on the same period and call the final version independent evidence.

Step 6: choose the next action

• Reject if coverage is weak, fires are too rare, or the outcome is fragile.
• Refine if one explicit assumption is wrong; create a new rule and artifact.
• Monitor if the evidence is coherent; run an alert before any execution workflow.

The correct output of a backtest is a decision about the rule, not an automatic trade.

For definitions of warmup, cooldown, signal density, and forward returns, use How to Read a Backtest Card. For more exercises, continue through Stingray Academy.

Still choosing a research setup? Compare no-code, Pine Script, and Python backtesting by data control, audit trail, monitoring, and maintenance work.