How to build a trading strategy with your own custom data

Jobless claims regime: SPY or Treasuries

Test window

Oct 2021 – Sep 2026

Starting balance

$100,000

Fee per trade

0.1%

Benchmark

SPY

This is the result. The four steps below show how the strategy was built and tested.

1. Describe the idea to Aurora

A prompt like this one describes the strategy in plain English.

“Build a portfolio that follows these rules. Data: an uploaded series of U.S. weekly initial jobless claims. Hold SPY while claims are below their value 252 periods earlier. Otherwise hold SHY (1-3 year Treasuries). Trade only when the regime flips. Backtest it between Oct 1, 2021 and Sep 30, 2026 with $100,000 to start, a 0.1% fee on each trade, SPY as the benchmark.”

2. Check the rules Aurora writes

These are the exact rules the backtest ran.

1. Rule 1

   Data: an uploaded series of U.S. weekly initial jobless claims.

2. Rule 2

   Hold SPY while claims are below their value 252 periods earlier.

3. Rule 3

   Otherwise hold SHY (1-3 year Treasuries).

4. Rule 4

   Trade only when the regime flips.

3. Backtest it against SPY after fees

The claims data I uploaded ends July 31, 2026. After that, the rule keeps using the last value.

Strategy return

+7.4%

SPY return

+88.5%

Largest fall

−28.9%

Sharpe ratio

0.17

Closed trades

42

Fees paid

$7,974

4. Test it on dates it never saw

Next, run a walk-forward test, which freezes the rules and checks them on later dates they never saw. See a walk-forward test on unseen dates

The results describe historical observations. They do not establish live trading performance.

Continue exploring