AQAI QuantAI research lab for systematic strategies

Forward portfolio

LIVE FORWARD · PAPER

The portfolio trades in real time through a paper brokerage account. Reported returns include the transaction costs, commissions and slippage applied by the portfolio's execution model. No real money or client capital is involved, so the results remain hypothetical.

Inception Mar 2026Active strategies 11Updated 2026-09-04Latest strategy added 2026-09-08
Total return
+12.0%
SPY +15.2%
Sharpe
3.15
SPY 2.10
Max drawdown
−2.4%
SPY −6.6%
Recovery time
39 days
SPY 10 days
Beta vs SPY
0.27
1.00 = market
Return at market β
+44.6%
SPY +15.2%
Max drawdown at market β
−7.0%
SPY −6.6%
Sharpe at market β
3.15
unchanged by scaling

The three market β tiles restate the record at the market's beta: the portfolio runs at β 0.27, so the same strategies held at 3.70× the exposure would carry the market's systematic risk. Return and the cumulative line stretch by exactly that factor and Sharpe does not move at all; the drawdown is re-measured on the stretched line rather than divided, because the peak it falls from is stretched too. It is arithmetic on the record above, not a second track record: no financing cost, borrow or margin is modelled, and nothing was traded at that size.

Cumulative return vs SPY

Solid blue: portfolio. Dashed: SPY. Violet: the portfolio at market β — its own line ÷ β. Fixed-base daily returns, summed — each line ends exactly at its total return above.

Drawdown

Peak-to-trough of the cumulative line, in units of starting capital. Recovery time above counts trading days from the deepest trough back to that peak.

Monthly returns

JanFebMarAprMayJunJulAugSepOctNovDecYear
2026+0.2%+7.4%+2.6%−0.1%+0.5%+1.5%+0.0%+12.0%
SPY−3.0%+10.2%+5.6%−1.3%+0.0%+3.0%+0.5%+15.0%

Portfolio above, SPY beneath it. Both rows are in units of starting capital, so each row's months add up to its own year and to the totals above.

Download daily returns (CSV) →

Composition

Strategy cluster mix

  • Event Driven28.6%2 strats
  • Sentiment Contrarian20.0%1 strat
  • Mean Reversion17.2%1 strat
  • Fundamental Event15.4%1 strat
  • Momentum10.2%3 strats
  • Event Sentiment6.7%1 strat
  • Sector Rotation1.0%1 strat
  • Trend Following1.0%1 strat

Capital weight by strategy family across the active book.

Subcluster mix

  • Sentiment Contrarian · Negative News Dispersion Reversal20.0%1 strat
  • Event Driven · Post-Report Cash Flow Drift20.0%1 strat
  • Mean Reversion · Oversold Multi-Day Pullback17.2%1 strat
  • Fundamental Event · Industry-Relative Ncfo Quantile Long-Short15.4%1 strat
  • Event Driven · Earnings/Event Drift8.6%1 strat
  • Event Sentiment · Earnings Risk-Language Shock6.7%1 strat
  • Momentum · Earnings/Event Drift4.6%1 strat
  • Momentum · Dual-Horizon Volatility-Scaled Momentum3.6%1 strat
  • Momentum · Dispersion-Filtered Cross-Sectional Momentum2.0%1 strat
  • Sector Rotation · Defensive Relative Strength Treasury Overlay1.0%1 strat

The same weights one level finer. Names describe mechanisms, not positions.

Sector exposure

  • Financial Services29.8%
  • Technology13.1%
  • Healthcare10.7%
  • Industrials10.3%
  • Consumer Cyclical9.0%
  • Consumer Defensive5.6%
  • Energy5.0%
  • Utilities4.2%
  • Real Estate4.2%
  • Communication Services4.1%
  • Other / unclassified4.0%

Each strategy's weight spread evenly across its tradable universe — universe-based exposure, not live position weights.

Individual strategies, signals and positions are not disclosed; composition is shown at the level of families and universes.

How the portfolio is run

Selection. Every candidate starts as a paper or a hypothesis, is implemented as code, audited (look-ahead checks, fidelity against the source paper) and backtested with modelled costs. Strategies are selected for forward testing editorially, on top of those stored results — a published backtest is not an inclusion decision, and rejected candidates stay published as research. A forward strategy that degrades is stopped; its history remains in the record.

Weights. Each live strategy trades a fixed capital allocation; the weights shown above are those allocations as a share of the book. Allocation studies use our optimizer toolkit: a mean-variance family (maximum Sharpe with zero risk-free, minimum variance, risk parity, custom mean-variance) solved with SLSQP under long-only, fully-invested constraints — weights sum to one, no shorting, no leverage — with Ledoit-Wolf shrinkage covariance, annualised from daily data. Positions under 0.1% are dropped and the rest renormalised. Rolling studies re-optimise monthly on an estimation window that advances one month per rebalance, require at least 20 trading days of data, fall back to equal weight when a window is too short or a solve fails, and can screen candidates on rolling Sharpe, annualised return and correlation to SPY.

Execution and costs. Orders route through a single dispatcher into a paper brokerage account (Interactive Brokers simulation); market-on-close is the default order type. Backtests apply per-trade commissions at IB rates and slippage/fee assumptions in basis points; the paper account applies the broker's own simulated commissions and market fills. Strategies run on weekdays, 08:00–17:30 ET.

Accounting. Each day's portfolio return is the capital-weighted sum of strategy returns, recorded at the close and never revised. Daily returns are measured against starting capital and summed, so the cumulative line and the monthly table add up exactly to the total return. SPY is compounded from its own daily returns and then expressed on that same starting-capital base, which is what makes the benchmark row of the monthly table comparable line by line and equal to the SPY total return. Recovery time counts trading days from the deepest trough of the cumulative line to the first close back at the peak that preceded it; while the line is still below that peak it counts the days so far. Taxes are not modelled.

Beta, and the figures at market beta. Beta is the ordinary least-squares slope of the portfolio's daily returns on SPY's, over every trading day the two have in common since inception — one number for the whole record, not a rolling one. The market β tiles and the violet line show what holding the same strategies at 1/β the exposure would do: the cumulative line and the total return are multiplied by that factor, Sharpe is untouched by it (mean and deviation scale together), and the drawdown is measured again on the scaled line rather than multiplied — a line that gave back 9% of a +10% peak gives back 17% of a +20% one, not 18%. They exist because a portfolio at β 0.3 and one at β 1.0 are not comparable to the index on their raw returns. They are a rescaling of the record and nothing more: no leverage was used, and no financing cost, borrow, margin call or the path-dependence of actually running at that size is modelled. Below β 0.10 the scaling is not shown at all — the implied exposure passes 10× and the result describes the noise in a beta estimate rather than the portfolio.

How results are labelled

ResultLabel
Historical simulation of a strategyBACKTEST
Paper-account trading on new data (this page)LIVE FORWARD · PAPER
Trading with real capitalLIVE CAPITAL — none yet; will be labelled separately