Systematic futures trading, measured in the open

Polarity Engine trades three of the major U.S. equity-index futures — YM (E-mini Dow), NQ (E-mini Nasdaq-100) and RTY (E-mini Russell 2000) — using a proprietary, fully rules-based trading system. Every entry, exit and position size is mechanical; nothing here depends on discretion or a forecast.

The system has been backtested on more than six years of minute-level market data per market, then held to a stricter bar than a standard backtest: each market’s rules were validated out-of-sample with walk-forward testing, frozen, and put into a live forward test that records every trade as it happens. What you see below is that record — updating continuously from the live program ledgers, reported as percentages of each program’s reference account and of the margin actually used.

The Polarity Engine sacrifices trade frequency in favor of selectivity, seeking a small number of high-conviction opportunities with strong net expectancy, limited execution friction, and no dependence on latency or specialized infrastructure.

Why low frequency

Across three index futures programs — Dow (YM), Nasdaq (NQ), and Russell (RTY) — the Polarity Engine has taken 289 trades in roughly six and a half years of history. That works out to about four a month with all three markets combined, and nearly half of all calendar months pass without a single trade. Those flat months are included in every ratio we publish; they lower the numbers rather than flatter them.

Most systematic shops would treat that as a weakness. We built for it deliberately, because nearly everything that makes fast trading expensive and fragile falls away when you slow down. There are no colocated servers, no proprietary exchange feeds, no latency arms race. Entries and exits key off the close of a four-hour bar, so the system runs on ordinary broker infrastructure, and a fill that arrives a few seconds late costs essentially nothing. Fewer trades also means less paid to the market in commissions, spread, and slippage — and every figure we show is net of modeled round-turn costs on every contract.

There’s a durability argument too. A speed edge erodes the moment someone builds something faster. The engine’s edge is a repeatable market behavior around price levels the market has never traded before, and every trade in the record can be pulled up and audited against the specific setup that fired it. Nothing about it depends on being first in line.

What selectivity has to earn

Trading rarely is only a virtue if the trades that do fire are unusually good. That’s the bargain the system makes, and it’s the bar we hold it to:

The validated record as published in each program’s trading plan. Live forward-test trades accrue on top of these counts and are reflected in the program cards below.

For the quantitatively minded: walk-forward efficiency is out-of-sample average profit per trade divided by in-sample, using parameters chosen only from data available at the time — a direct measure of how much of the edge survives outside the fitting window. Anything above 0.4 is generally considered robust. RTY’s 1.07 means it performed better out of sample than in. On risk-adjusted returns, the YM program runs a 1.80 Sharpe and 3.97 Sortino on monthly net P&L with a worst drawdown of 15.9%, recovered in 153 days; NQ runs a 1.77 Sharpe with a worst drawdown of 8.9%. These ratios are computed with every flat month included.

The RTY result deserves its own sentence. The structure was frozen from the YM playbook, the geometry declared in writing before the first run, and the test executed once with no re-fitting allowed: 69 trades, 94.2% win rate, profit factor 3.0. A blind pass on an untouched market is the strongest answer we can give to the question every allocator should ask — is this curve-fit?

Why this record isn’t curve-fit

Anyone can produce a backtest with impressive numbers. Tweak enough parameters and you can always find settings that would have made money on historical data — the results look good because the rules were fitted to the past, not because they capture a real edge. That’s overfitting, or curve-fitting, and it’s why sophisticated allocators treat standalone backtest results as nearly worthless. We designed our process specifically to close that door:

• Rules registered before running. Every rule and every parameter is committed to a version-controlled record before the first simulation, so nothing can be quietly tuned to fit the outcome.
• Blind first pass. The strategy runs once at the registered settings, untouched, and the result stands whatever it says. The RTY record above is exactly such a run.
• Walk-forward validation. The rules are graded on data they never saw, fold by fold — that’s the walk-forward efficiency figure in the table above.
• Frozen, then forward-tested. After validation the parameters lock, and a live paper-trading engine records every trade as it happens. A forward test can’t be curve-fit — the future hasn’t happened yet. Changing any frozen parameter restarts the clock.

The failures get the same treatment as the successes. When ES failed its walk-forward, the program stopped and the failure was published. When the breakthrough component failed structurally on NQ, it was dropped rather than re-tuned.

We’re candid about the remaining trade-off: 289 trades is a small sample by high-frequency standards, and it always will be — that’s inherent to selectivity. It’s exactly why the walk-forward results, the blind RTY test, and the live forward ledgers carry more weight here than any additional in-sample statistic. Every position carries a hard stop from the moment of entry, sizing is fixed by rule, and the system’s worst historical day is disclosed, not averaged away.

Nearly half our months are flat, and every one of them is the system working: no forced trades, no costs bleeding out, capital intact and waiting. That patience has carried the engine through a pandemic crash, the fastest rate-hiking cycle in four decades, a bear market, and two historic bull runs — regimes changed; the rules never had to. And the record isn’t finished: the trade count on this page is live, and the program cards below are the forward ledgers writing the next chapter in real time. Most track records ask you to trust the past. This one invites you to watch.