Shorting a bear market doesn't work: the cross-asset verdict that surprised us

The intuition everyone shares

When the market collapses, the intuition is irresistible: you just had to short it. In hindsight, every bear market looks like a highway pointing down. We wanted to know whether that highway exists before you know what comes next — and we had the tooling to test it seriously.

The protocol: forge on one market, judge on another

A two-step method, designed to prevent hindsight cheating:

1. Forge: massive GPU search of strategies (including short variants) on a bear phase of Bitcoin — thousands of parameter combinations, selection of the best.
2. Cross-asset hold-out: the champions are then evaluated, with zero additional tuning, on a bear phase of Ethereum — a market they had never seen in any form.

If "short the bear" is a real edge, it must survive this transfer.

The verdict

• Only about 2% of short strategies beat buy & hold on the transfer market — a rate consistent with plain selection luck.
• The short champion did worse than holding the position: relief rallies, violent and frequent in bear markets, liquidate shorts before the trend resumes. A bear market doesn't go down in a straight line — it goes down in an elevator and back up in a rocket, repeatedly.
• What actually transferred, to our own surprise, was defense: long-only strategies with cash exits. On the never-seen market, nearly the entire top of the survivors was defensive, with a significant margin above buy & hold — not by winning more, but by losing far less.

Plainly: in a falling market, the winning move wasn't betting on the fall. It was not playing.

(These results come from our GPU bear forge campaign, BTC→ETH hold-out; orders of magnitude are reported as measured on that single campaign — one market pair is a sample, not a law.)

Why this fits everything else we measured

This result completes our World Arena matrix: in rising markets, 0% of our strategies beat buy & hold; in falling markets, shorts get crushed and only defense survives the transfer. Both times the lesson is the same: activity must prove it pays, and the only admissible proof comes from data the strategy has never seen.

And there is an uncomfortable symmetry: the defense that shines in a bear underperforms in a bull — consistent with the matrix. Anyone selling you an "all-weather" strategy without showing per-regime behavior is selling you half the picture.

Test your own intuition before it charges you for the lesson

• Free Windows Lab: replay this kind of protocol locally — search on one slice, verdict on a sealed holdout the search never saw. Educational simulation, no profit promises.
• Free Quick Check: paste your Pine script, no account. The independent audit locks data, protocol and engine before any payment — and an unfavorable verdict is still a delivered audit. Nobody can buy a better verdict.