Performance & Research
Performance
Live results come first, shown on a percentage basis and indexed to 100 at inception. Capital amounts and account values are not disclosed. The historical research behind the strategy follows further down, clearly marked as hypothetical.
Below the live resultsExplore the research behind the strategyHistorical backtest, how it was tested, how it trades, and where it can go wrong.Live results
Live · Real account- Inception
- 19 October 2026
- Data as of
- 6 November 2026
- Trading sessions
- 15
- Since inception
- −15.6%
- SPY −2.9%QQQ −6.9%
- To 6 November 2026
- Best week
- +0.1%
- 3 completed weeks, to 6 November 2026
- Worst week
- −8.8%
- 3 completed weeks, to 6 November 2026
- Max. weekly drawdown
- −15.7%
- Peak to trough on Friday closes
Weekly figures use Friday-to-Friday returns of completed weeks (Monday to Friday; if Friday is a market holiday, the last session of that week), starting from the close before inception. A week in progress is not counted. Maximum weekly drawdown is the largest decline of the index from a previous weekly high, measured on week-end closes only, so moves within a week are not included. CAGR and Sharpe ratio are shown once 12 months of live history exist.
Cumulative performance
Indexed to 100 at the close of 16 October 2026, the session before inception.
Returns by period
| Period | RODEO | SPY | QQQ |
|---|---|---|---|
| Since inception | −15.62% | −2.92% | −6.85% |
| Last 5 sessions | −8.78% | −2.64% | −4.92% |
| Latest day | +4.05% | +0.92% | +2.02% |
Trailing periods in trading sessions, ending 6 November 2026; they need not align with calendar weeks. Periods are shown only once enough live history exists.
Methodology
- Returns are daily and time-weighted. Deposits and withdrawals are not performance: each external flow is removed from the day's gain and added to that day's starting capital, because it is credited before the session in which it is invested. Returns are net of trading commissions and currency-conversion costs, and measured in EUR.
- SPY and QQQ are total returns (dividends reinvested) converted to EUR over the same sessions, as reference benchmarks.
- Once shown, the Sharpe ratio is annualised from daily returns with a 2.0% annual risk-free rate.
The live section contains live results only; nothing from the historical research below is included in it. Past performance is not indicative of future results. This information is provided for informational purposes only and does not constitute investment advice, an offer, or a solicitation to buy or sell any security or financial product.
Historical research · Hypothetical
The research behind the strategy
The live RODEO portfolio is a systematic, long-only strategy in large, liquid U.S. stocks. Each week it ranks companies on a small set of quantitative signals and holds a concentrated portfolio of the most attractive ones. This section shows how that approach looked in historical testing, and how it works, without giving away the recipe: exact formulas, weights and trading rules are kept private.
Everything in this section is hypothetical: simulations of the rules on historical data, not returns earned by an actual account. It is kept separate from live performance and is never combined with it.
The evidence comes in three layers, and each deserves a different amount of trust.
Historical research
January 2008 – August 2025 · in-sample
19.3%
a year (CAGR), vs. 11.1% for SPY
The longest record, but the design was chosen on this same data, so it flatters the strategy.
Pre-registered holdout
January 2024 – September 2025 · 87 weekly decisions
+124%
cumulative over the whole period, not per year; SPY +39%
Data unseen during development, opened once. Short, and in a favourable market.
Live trading
Since October 2026 · real account
The only record that involves real money, real fills and no hindsight. It is still very short and should not be read as indicative yet.
Back to live results1 · Historical research sampleHypothetical · Backtest
Growth of 100, January 2008 – August 2025
Logarithmic scale, so that equal percentage moves look equal across the whole period. Week-end values.
Drawdown from previous peak
Deepest daily decline within each week.
Summary statistics
| Measure | Research | SPY |
|---|---|---|
| CAGR | 19.3% | 11.1% |
| Sharpe ratio | 0.67 | 0.56 |
| Maximum drawdown | −65.0% | −51.5% |
| Volatilityannualised | 32% | 20% |
| Walk-forward CAGRdesign re-chosen each year with earlier data only | 16.5% | — |
| Walk-forward Sharpe ratio | 0.59 | — |
The walk-forward rows are the more realistic guide: when a key design choice was re-made each year using only data available at the time, results were lower than for the version chosen with hindsight.
Returns varied a lot by period
| Sub-period | Bars | Research | SPY |
|---|---|---|---|
| 2008 – 2015 | 2.5% | 6.9% | |
| 2016 – 2019 | 19.9% | 14.3% | |
| 2020 – 2023 | 40.8% | 12.0% | |
| Jan 2024 – Aug 2025 | 64.5% | 21.6% |
Annualised (CAGR). Bars share one linear scale starting at zero. Red marks a sub-period in which the research strategy trailed SPY.
2 · Out-of-sample holdoutHypothetical · Holdout
January 2024 – September 2025
| Measure | Research | SPY |
|---|---|---|
| Cumulative returnwhole period, not annualised | +124% | +39% |
| Sharpe ratio | 1.42 | 0.96 |
| Maximum drawdown | −26.4% | −18.8% |
The rules and the pass/fail criteria were fixed in writing before this period was examined, and the test was run once. It passed those criteria.
It is also a short period that suited the strategy. Five stocks accounted for about 57% of the gains, and without the three best weeks the cumulative return falls from +124% to +44%. The strategy carried more market risk than SPY (a beta of about 1.4). After adjusting for market, size, value and trend exposure, the estimated alpha was positive but not statistically significant (t-statistic about 1.1): 87 weeks is too short to tell skill from a favourable market.
Net of modelled commissions and bid-ask spreads, weekly rebalancing, U.S. dollar returns, point-in-time historical research data. Sharpe ratios use the daily excess return over the risk-free rate, annualised. SPY total return over identical dates. Research CAGR is annualised over trading sessions. Backtests are simulations: they assume execution at modelled prices and costs and are subject to model and data limitations.
Methodology · Economic intuition
The idea
The strategy rests on two ideas from empirical asset pricing. Neither is new on its own; the research question was whether they are more useful together, and whether the combination survives real trading costs.
01
What option prices reveal
An option price contains the market’s own estimate of how much a stock is likely to move. Comparing that forward-looking estimate with how much the stock has actually been moving says something about how nervous, or how relaxed, investors are about that particular company. The research asks whether that gap carries information about returns that the share price has not yet absorbed.
02
Trends that persist
Stocks that have done well relative to their peers over the preceding months have, on average, tended to keep doing so for a while. It is one of the most widely documented patterns in financial research, usually attributed to investors adjusting gradually to new information.
In testing, neither idea was enough by itself. The option-market signal on its own did not hold up, and the trend signal alone was considerably weaker than the combination. The most informative cases were stocks where both pointed the same way, and that interaction was strongest in turbulent markets.
Several other ideas were researched along the way and set aside, including signals based on insider trading activity, earnings announcements, the shape of option prices, value and quality measures, and switching between market regimes. None of them is part of the live strategy.
Strategy behaviour
How it trades
Positions are chosen systematically and change when their relative attractiveness changes, not because of a view on the news, the market or a particular company.
Step 1
Observe
After the last session of each week, the strategy looks at a broad universe of large, liquid U.S. companies, using only information available at that close.
Step 2
Rank
Every eligible company is scored on the signals and ranked against the others. Nothing about an individual company is decided by hand.
Step 3
Compare
Current holdings are kept while they remain among the more attractive names. A position is replaced when its ranking has fallen far enough behind; the gap between ‘good enough to buy’ and ‘weak enough to sell’ avoids trading on noise.
Step 4
Trade
Any changes are made in the following session. New positions are sized so that no single view dominates the portfolio when it enters.
- Typical trades
- ~2.7 / week
- Median holding
- ~5 weeks
- Modelled costs
- ~3.8 pp / year
Aggregate figures from the hypothetical research sample (January 2008 – August 2025). Costs are commissions and bid-ask spreads, expressed as the reduction in annual return. In the holdout period the strategy made about 3.1 trades a week with a median holding of 4 weeks. Current holdings and individual trades are not published.
Method
From research to live trading
A strategy that only works on the data it was designed on is not worth much. Most of the effort went into trying to break it, and into making sure that what was tested is what is actually traded.
01
Hypothesis
Start from an economic reason a signal might work, and write down what would count as evidence.
02
Historical testing
Test on point-in-time data from 2008 onwards, after modelled commissions and bid-ask spreads, across sub-periods and market conditions. The most recent data is locked away and not looked at.
03
Rules frozen
Before opening the locked period, the full set of rules and the criteria for passing or failing are written down and fixed.
04
Holdout test
The locked 2024–2025 data is opened once, for that frozen specification only. No rule may change because of what it shows.
05
Robustness
Walk-forward selection, deliberately noisy inputs, stress periods, and a rebuild of the signals from the broker’s own market data instead of the research datasets.
06
Paper trading
The exact rules run on a paper account to check data, orders and reconciliation end to end.
07
Live
Real-money trading since October 2026, reported in the live section at the top of this page.
One honest caveat: after the holdout test, one weighting parameter was adjusted using data that includes the holdout period. The live strategy is therefore not identical to the version that took the holdout test, and the holdout is not out-of-sample for the live version itself.
Implementation
How live trading differs from the backtest
Real prices, real costs
The backtest trades at modelled prices with modelled commissions and spreads. Live trades are entered by hand at the broker and pay whatever the market and the broker actually charge.
Different data
Live signals are built from the broker’s market data rather than the research datasets. When the research period from November 2016 to August 2025 was rebuilt that way, weekly returns were highly correlated (about 0.94), but only around 66% of holdings matched at any time: small data differences change which stocks get picked.
Currency
The research is measured in U.S. dollars. The live account is measured in euros, so exchange-rate moves and conversion costs are part of live results.
No hindsight
The backtest was run knowing how the story ended, and its design was chosen with that knowledge. Live decisions are made one week at a time, with no second attempts.
Practical frictions
A small cash buffer is held, trades happen during real sessions rather than at an idealised close, and operational issues such as data gaps or delayed orders can occur.
Read this too
Limitations and risks
The historical numbers are encouraging, and they come with serious caveats. These are the ones that matter most.
- 01
Deep drawdowns. In the research sample the strategy fell 65% from its peak, and lost 51% in 2008 alone. Losses of that size can take years to recover.
- 02
Long weak stretches. From 2008 to 2015 the research strategy earned 2.5% a year, behind SPY’s 6.9%. Most of the historical gain came after 2016, and especially from 2020 onwards.
- 03
More volatile than the market. Annualised volatility was about 32% against 20% for SPY, with a beta of about 1.1. In ordinary sell-offs it has tended to fall further than the market.
- 04
Concentration. The portfolio holds few stocks, and a handful of large winners and a few exceptional weeks account for a large share of historical gains.
- 05
Hindsight in the design. Parameters were selected on the same history. The walk-forward test and a test with deliberately noisy option data (Sharpe ratio 0.55–0.59 instead of 0.67) both suggest the headline figures overstate what to expect.
- 06
Costs. Turnover is high, so costs matter; the research models them, but real costs can differ.
- 07
Signals can stop working. Patterns that held in the past can weaken as markets change or as more investors exploit them.
- 08
A short live record. Live results so far cover weeks, not years, and say little about long-run behaviour.
Important. Backtested and historical research results are hypothetical. They were produced by applying rules to historical data with the benefit of hindsight, do not represent trading in an actual account, and do not reflect the impact that market conditions, liquidity or decisions made at the time might have had. Hypothetical results have inherent limitations, and no representation is made that any account will or is likely to achieve similar results. Live results are shown separately in the live section at the top of this page. Past performance, actual or hypothetical, is not indicative of future results. Nothing on this page is investment advice, an offer, or a solicitation to buy or sell any security or financial product.
Questions about the research are welcome: inforodeoinv@gmail.com