Did short positions lose money in the strategy tournament?

That's not the comparison being made. The question was whether adding short positions improved on the same strategy's own long-only version, and it only did in 2 of 8 cases in validation and 3 of 8 in test. In most cases the short version lagged its own long-only counterpart.

Who actually pays funding on a perpetual?

It follows the perpetual's price premium over the spot index, not who's in the market's minority or majority. With positive funding, the historical norm, longs pay and shorts collect — the opposite of the intuition that betting against the trend is what costs money.

How much did this experiment's shorts actually pay or collect?

Measured conditioned on the real position in validation: on their short days, the two strategies that reached a real short version COLLECTED between 0.006% and 0.012% daily (roughly 2-4% annualized). On their long days, the perpetual-financed trading track PAID between 0.060% and 0.089% daily (roughly 22-32% annualized), the period's real financing cost, carried by the long/short package itself, not by the shorts.

Did the equity and gold shorts follow the same mechanism as the crypto ones?

Not exactly. The S&P 500 short was simulated on the total-return index, so it pays dividends by construction, which is its dominant real cost. The gold short uses a futures contract, where that cost is already embedded in the price. Neither carries an explicit daily funding rate the way crypto perpetuals do.

Does this mean no strategy with short positions can ever work?

No — it means that in this specific experiment, with the real funding mechanism applied to both legs, none managed it consistently. If this research ever became a product, it would be long/cash, never long/short, precisely because of this result.

Two glowing arrows forming a loop between a cluster of rising green candlesticks and a cluster of falling red candlesticks, with a Bitcoin coin where they meet — the arrow flowing out of the green (long) side is thicker and brighter than the one returning from the red (short) side — Original illustration generated with AI (Adobe Firefly) — not a representation of real data.

The Short Side Loses Even While Collecting Funding

NodeWitness Team August 14, 2026 5 min read

In the 2,255-strategy tournament, adding short positions only improved on the same strategy's long-only version in 2 of 8 cases in validation and 3 of 8 in test. It wasn't about financing: on their short days, the strategies actually COLLECTED funding, and still lost on direction. The real financing cost landed on the long/short package's own long legs, not on the shorts.

Winning on the way down sounds like the missing half of buy-and-hold. The strategy tournament covered earlier in this series tested that idea against real funding data, not a convenient estimate, and the result was more surprising than expected: the shorts lost even on days when funding itself was paying them. This piece breaks down who really pays that financing, and why the full package, long and short together, can't beat its own long-only version.

The cost map of a short position

Real funding: who pays, who collects

The figure already familiar from this series, an average of 0.0321% daily (roughly 11.7% annualized) over Binance's full funding history for the period, is the period's average, but on its own it doesn't say who actually paid it. Measured conditioned on the real position of the two strategies that reached a genuine short version: on their short days, they COLLECTED a small subsidy, between 0.006% and 0.012% daily depending on asset and strategy (roughly 2-4% annualized). On their long days, by contrast, the perpetual-financed trading track PAID between 0.060% and 0.089% daily (roughly 22-32% annualized), the real cost their long-only spot twin never bears, because spot carries no funding. The 10-30% annual range the experiment's own design had pre-registered before computing anything held up exactly, just landing on the package's long legs, not on its shorts.

Two horizontal bars showing how often the short version beat its own long-only version: 2 of 8 (25%) in validation, 3 of 8 (37.5%) in test, both below half the cases, with a note on real Bitcoin funding over the period (~11.7% annualized) and who pays it when funding is positive — Original diagram built from the experiment's own verified CSV output: the short side stayed below half the cases in both partitions.

The result: 2 of 8 in validation, 3 of 8 in test

Of the two strategy families that reached a real short-position version, neither managed to make that version consistently beat its own long-only counterpart. Counting each asset-strategy combination separately (eight per partition), the short won in 2 of 8 cases in validation and 3 of 8 in test. Not close to half, and the same asset never won in both partitions at once: whatever edge appeared didn't survive from one partition to the next.

Why the short side loses even while collecting

This experiment's shorts lost on direction, not on financing: even while collecting a small subsidy on their short days, price movement beat them more often than they beat it. And the full package, long and short combined, carries an extra cost its long-only spot version never pays: the real funding on its own perpetual long legs, running 22% to 32% annualized. Two separate reasons to lose, not one. The equity and gold shorts don't carry that same daily funding (the total-return index already prices in dividends, and the gold futures contract has carry embedded in price), but neither escaped the general pattern: on no asset did the short side demonstrate a consistent edge over its own long-only version.

What this means for any leveraged short strategy

This result doesn't say winning on the downside is impossible. It says that, with the real funding mechanism applied to both legs and no favorable number invented, doing it systematically with financed short positions is far harder than looking at price direction alone would suggest. If this tournament ever moved from experiment to product, the conclusion is already settled: long/cash, never long/short. Same discipline that led to "zero clearly effective strategies" in the tournament's overall verdict: a negative result, measured with the same rigor as a positive one, and published just the same.

There's also a cost this experiment doesn't measure because its simulation doesn't carry it: liquidation risk. A leveraged short on a perpetual can get force-closed on a sharp enough upward wick even if the position would have eventually been proven right, which turns a temporary drawdown into a permanent, realized loss at the worst possible moment. The backtest's funding numbers are the visible cost. That structural asymmetry between a leveraged short and a spot long, which simply can't be liquidated the same way, is a real cost too, just one that doesn't show up cleanly in a funding-rate average, and it would only make the short side's already weak record in this experiment look worse, not better, if it were added on top.

Nothing in this article is investment advice, and nothing on NodeWitness is. You can read the tournament's full verdict, why backtests lie even the careful ones, or follow NodeWitness's live Cycle Score, same discipline of publishing what doesn't work too.

Last updated: August 31, 2026