Is your capital working, or is it just you?EdgeQuant, your capital’s wake-up call.

• Already trading crypto?
  Then balance your trading with Nasdaq and gold.
• Taking prop challenges?
  Stop repeating judgment calls. Repeat a system.
• Have capital but no time?
  The rules stay the same, and the app places the orders.

You decide how much to automate.

Every tier gets the same signals.

Times shown in US Eastern time.

Each asset entered the portfolio at a different time, so cumulative R cannot be compared directly between them.

Day to day the three barely move together: daily P&L correlations are NQ-GC -0.00, NQ-BTC never on the same day, GC-BTC never on the same day (|r| ≤ 0.00, 1,891 calendar days since 2021-07, no-trade days counted as zero). Drawdowns shrink too: over the window since 2021-07, when all three ran together, each asset’s worst run adds up to 50.1R, while the combined portfolio’s is 22.6R.

You can start with one. The three-market record above is a destination, not a requirement to get started. Accounts connect per asset, and the system works with just one connected. You still receive signals for all three and choose which ones to act on. Without your own trading capital, you can trade gold and Nasdaq at standard size through the prop route. To start with less, you can trade fractional bitcoin on a crypto exchange, with no contract minimum.

• 01PredictionInstead of combining hundreds of indicators, a learned model reads a small set of decision-relevant inputs and predicts the market direction. The same inputs always produce the same result.
• 02SizeA correct direction does not mean the same position size. Position size is set per signal, between 1R and 3R. Apart from no-trade days announced in advance, early closes and market holidays, there is no human discretion in the loop.
• 03Per-asset filtersEach asset has its own filters. When a filter blocks a trade, the card says NO TRADE.

The three layers, the features and the validation pipeline are set out in Algorithm architecture.

Long, short, or no trade by design. Every member gets the same card, showing the market, direction, entry reference, stop, exit time, and whether the confidence and risk filters passed. If conditions are not met, it says NO TRADE and explains why. The system records that decision on the same terms as a trade.

The R figure next to the direction is relative position size, not an amount of money. Confidence is the gate that decides whether to enter at all; it does not set the size.

From there, execution is handled by EQ Autopilot. ↓

The paid tiers are monthly subscriptions, and the price shown includes value added tax.
Cancel any time; a 14-day right of withdrawal applies.
The Autopilot tier, including its trial, is not available to US residents (Preview and Operator are).
For restrictions in your country, see Section 7 of the Terms.

The amount on the Autopilot card is the regular price. Right now the founding member rate of $199/month is running, and how many places remain at each rate is published on the membership page.

Walk-forward validation trains the model on past data, then tests it on the next period the model has not seen. We repeat this process across successive periods. The figures here come from those out-of-sample tests.

This describes how the model learns. Other design choices - the session each asset trades, when it entered the portfolio, the execution rules (the conditions that filter out entries, and exits), the risk multipliers, and some learning settings - were set after reviewing the full period. We chose settings that held up better under stress tests, even when they produced lower returns.

All figures are net of exchange fees, commissions and slippage, and exclude the EdgeQuant subscription fee. Entries, timed exits and stops are modelled as market fills; we assume no savings from limit orders. Slippage is set at half a tick per futures leg and 0.002% per crypto leg. See the robustness tests for the two-tick stress case.

How long the deepest drawdown lasted matters too. The peak formed in April 2016, the trough came in November 2017, and the new high arrived in January 2018 - 21 months from peak to new high. Using this system means being able to hold the same rules through 21 months without a new high. (The max drawdown measures the full decline across two calendar years, which is why it is larger than any single worst year - the two numbers measure different windows.)

Reading these figures: 2024 contributed +248R, about 34% of the total cumulative result; excluding it, the Recovery Factor falls from 17.48 to 11.6. Since July 2021, when all three assets have run together, the result is +538R with a max drawdown of 22.6R. Over the last 22 months (since January 2025), bitcoin made up about 69% of the result - Nasdaq and gold +44.9R, bitcoin +101.5R. The 42.2R max drawdown ran from April 2016 to November 2017 and was recovered by January 2018 (21 months from peak to new high), before Bitcoin joined the portfolio. The first and current years are partial, and a dash (-) means the asset was not yet in the portfolio. We do not set a minimum number of signals per day, week or month - a stretch of several days without a signal is normal system behavior, not a defect.

Without 2024, the total is +489.1R, the max drawdown 42.2R and the Recovery Factor 11.6. 2024 alone is about 34% of the total. To see the result without leaning on one year, look at this curve.

See the test methods on one page →

Equal-weight portfolio of NQ, Gold, Bitcoin - record from October 2014 to October 2026.

Before the returns: how far did it fall?

Same $115,000, same period (since April 2017). Our model is net of trading costs (commissions and slippage) and fees.

Top: month-end cumulative P&L, which changes places depending on when you look. Bottom: each month's deepest drawdown (closer to zero is shallower). DCA is measured from the peak of the account value (money put in included), our model from the peak of the account net of fees.

Both sides use the same $115,000 in total, on data since April 2017, but it goes in differently: DCA spreads it over the months, our model invests it all on day one. Plain DCA puts in $1,000 a month - daily, weekly or monthly, each shown separately - regardless of signals; signal-timed DCA spends the same monthly budget only on days with a long signal. All DCA plans use the mix a weight sweep ranked best by return per unit of drawdown, Nasdaq 70% / gold 30%. Our model puts the full amount in on day one, split equally across its three assets, re-sizes 1R to 0.4% of current capital on every trade (compounding), and pays the Autopilot tier's list price of $399 every month.

Our model's worst drawdown was −10.7% (July 2020) - far smaller than DCA's: every DCA plan fell −32.5% or more (October 2022).

Return divided by worst drawdown is 101.8× for our model against 4.7× for the best DCA plan - about 21.5× apart. Accepting the same 32.6% drawdown as DCA, our model could carry about 3.6× the risk, and the result would be far above DCA - we do not recommend that level.

Total P&L was +$1,257,613 for our model net of $45,885 in fees, and +$177,665 for the best DCA plan. Cumulative P&L changes places depending on when you look (DCA was ahead for 8% of this period).

For members on the 20% standing discount (after passing a 150K evaluation), fees over the period come to $36,708 and our model's P&L to +$1,308,326.

Both sides had the same three assets available. The weight sweep, which ranks mixes by return per unit of drawdown, drove bitcoin’s weight in a DCA plan to zero: a plan that only buys and never sells has no way to cut its drawdowns. It stays in our row because our model can take the short side and stop out.

Sizing basis: the model row uses the recommended 0.4% risk per trade and re-sizes 1R to current capital on every trade - it starts at 1R = $460 on $115,000 (a $76,667 futures share for Nasdaq and gold, and the same allocation for bitcoin). In EQ Autopilot you keep the account’s 1R in line with its capital. Higher risk raises profit and drawdown together.

Using the signals only to time purchases changes almost nothing.

All results are simulation; our model row is walk-forward, as of 2026-10-16. Three conditions are set in favour of DCA: its 70/30 mix was chosen with hindsight, its costs are assumed zero, and the comparison above uses whichever DCA cadence earned the most, together with the shallowest drawdown among them. Our model row includes measured fees, 0.5-tick slippage and the Autopilot tier's list price every month ($45,885 over 115 months). BTC perpetual funding fees are not deducted. The subscription did not exist for most of this period, so that cost is an assumption; taxes are left out on both sides. Drawdown is the fall from a peak as a share of the account, with deposits and fees treated as cash flows. DCA uses the same continuous unadjusted futures series the model trades (close prices, no dividends, no roll adjustment). The model drawdown here is for the period since 2017-04; the worst drawdown over the full history is 42.2R, about 17% of capital at the recommended risk, on the three-asset portfolio basis, from April 2016 to November 2017 (before Bitcoin joined). The $500 unit used elsewhere on this page is the same 0.4% on a $125,000 account. Past simulation results do not guarantee future performance.

Full analysis here →

The binding terms are in the Terms of Service: dormancy in Section 10, cancellation in Section 11, withdrawal in Section 12.
You can cancel from your dashboard or on the cancellation page. For anything else, write to support@edgequant.app.

The architect holds a Ph.D. in ECE from UC San Diego and works as a data scientist and researcher. The system follows the same research principles: reproducible results, walk-forward validation, and regular checks for ways a result could be wrong. The architect handles the model, execution, verification and record-keeping, and runs the same Signals in their own accounts. The results are public, and the architect is accountable for the system’s design and operation.

The name and business details are published in the legal notice (Impressum), and the research record and academic activity can be found on Google Scholar.