A Hunter selection is a snapshot. It is the portfolio that satisfied a set of rules over one window of past data. The obvious follow-up question is what happens to that snapshot afterwards, and whether running the same rules again later gives a different answer worth acting on. We tested that on our own saved settings.
Running the same Hunter rules again on a later date, using only data available up to that date, and switching to the newly returned portfolio. The rules stay fixed; only the data window moves.
What did re-picking show against holding one selection?
Re-picking monthly beat holding on both profit and worst drawdown in all four settings. Re-picking weekly added a little more profit in all four, with a mixed drawdown result.
| Setting | Pick once and hold | Re-pick monthly | Re-pick weekly |
|---|---|---|---|
| A: 52-week reference, $100k | +$69,815 / $20,309 | +$116,626 / $14,180 | +$126,270 / $13,097 |
| B: total-history reference, $100k | −$2,311 / $43,247 (from Feb 2025) | +$57,462 / $21,398 | +$63,042 / $17,765 |
| C: 26-week reference, $35k, confluence filter off | +$14,426 / $14,492 | +$44,190 / $11,317 | +$46,641 / $12,284 |
| D: 26-week reference, $35k | +$6,736 / $25,695 | +$57,478 / $8,473 | +$60,409 / $7,847 |
Each cell is profit / worst drawdown in US dollars. Source: CashFlow Engine backtest data, October 2023 to 9 October 2026 (about 36 months; setting B about 21 months). Hypothetical results, not live trading.
Three details matter when reading the table:
- The "pick once and hold" column is one start date, the first one. It is a single path, not an average.
- Setting B's total-history reference only exists for past dates from January 2025. The search on 3 January 2025 found no portfolio, so that run held cash for the month. Its hold baseline starts in February 2025, and the weekly run's hold baseline (from 10 January 2025) was +$5,656 / $32,969.
- Setting C carries a minimum confluence-score filter in its saved form. Confluence is only stored for the current day, so the past-date test ran without that filter.
Weekly versus monthly is less one-sided than monthly versus hold. Weekly earned more than monthly in all four settings, by $2,451 to $9,644. Its worst drawdown was smaller than monthly's in three of four and larger in setting C ($12,284 against $11,317).
How was the re-picking test run?
Every selection was made as of a past Friday using only data up to that Friday. The rules in each saved setting stayed fixed. Only the date moved.
For each setting we compared three ways of trading through the same period:
- Pick once and hold: run the rules as of one past Friday and keep that portfolio unchanged to the end.
- Re-pick monthly: on the first Friday of each month, run the same rules as of that Friday and switch to the new portfolio.
- Re-pick weekly: the same, every Friday.
The out-of-sample profit and loss of each held period was chained into one equity curve. Positions were one contract per selected strategy unless the Hunter assigned more. The profit and loss series are the same per-strategy daily results the Hunter itself works from.
Two further choices were made to keep the comparison honest:
- A reproducible search mode. For this study we used an internal search mode in which the same input always returns the same portfolio. Two complete runs matched on 37 of 37 months. The live Hunter limits each search by clock time, so a re-run there can return a different, near-equally-scored portfolio.
- Many start dates, not one. For every start month (and week) we held that start's selection to the end and compared it with re-picking from the same start. That produced 37 monthly and 158 weekly head-to-head comparisons per setting.
The four settings are the operator's own saved configurations, used here as a demonstration of the tool. They are not recommendations.
| Setting | Reference period | Budget | Portfolio size | Drawdown cap |
|---|---|---|---|---|
| A | 52 weeks | $100k | 10-18 | 7% |
| B | Total history | $100k | 10-20 | 12% |
| C | 26 weeks | $35k | 7-20 | 9% |
| D | 26 weeks | $35k | 7-15 | 12% |
How often did re-picking beat holding across start dates?
Across start dates with at least 12 months left to run, re-picking won most head-to-heads on profit in every setting. The drawdown record was strong for monthly in all four, and weaker for weekly in setting C.
| Setting | Monthly: more profit | Monthly: smaller drawdown | Weekly: more profit | Weekly: smaller drawdown |
|---|---|---|---|---|
| A | 18 of 25 | 25 of 25 | 91 of 106 | 97 of 106 |
| B | 7 of 9 | 7 of 9 | 38 of 40 | 39 of 40 |
| C | 20 of 25 | 19 of 25 | 87 of 106 | 51 of 106 |
| D | 22 of 25 | 23 of 25 | 103 of 106 | 100 of 106 |
Source: CashFlow Engine backtest data, October 2023 to 9 October 2026. Hypothetical results.
These counts overlap heavily. A start date one week later shares almost its entire path with its neighbor, so 106 weekly comparisons are far fewer than 106 independent observations. Setting B has only 9 monthly and 40 weekly comparisons because its history is short. Read the counts as consistency across starting points, not as a statistical significance claim.
Why might re-picking have helped? (Interpretation, not a proven cause)
One plausible reading is that a selection fitted to one window goes stale as market regimes shift. The test does not prove this mechanism.
The Hunter optimizes its rules on a past reference window. A portfolio that fits one stretch of market behavior well has no built-in reason to keep fitting the stretch after it. If conditions move, the strategies that earned the most in the old window can stop being the ones that earn in the new one. Running the same rules again on fresher data moves the selection toward whatever the recent window rewarded.
That is an interpretation. We did not isolate regime change as the cause, and other things differ between a held and a re-picked portfolio: the re-picked one also changes its strategy mix, contract counts and total exposure each time. The data shows the outcome. It does not show the reason.
Does a weekly re-pick change the portfolio too much?
A weekly re-pick swapped 65-78% of the strategies each week, and a monthly re-pick swapped about 90% each month. Turnover was large at both cadences.
A monthly swap rate that high does not mean the selection was unstable in a bad way: a month is a long gap, and the refreshed window had moved a lot. But it does mean the "re-picked" portfolio was close to a new portfolio each time rather than a gently adjusted old one. Each swap in practice would mean opening, closing or resizing positions, with whatever execution cost that carries. The backtest applied no such cost to the switching itself, so the profit figures for both cadences are before the practical friction of that turnover.
Get the weekly 0DTE research email
Research notes and product updates, straight to your inbox. Unsubscribe anytime.
How far did out-of-sample drawdowns run past the cap?
Out of sample, worst drawdowns ran about 1.5 to 4 times the cap set in the search: roughly double on three settings and close to four times on one. The cap constrained the in-sample reference window, not what happened afterwards.
| Setting | Cap in the search | Worst drawdown, monthly (% of budget) | Worst drawdown, weekly (% of budget) |
|---|---|---|---|
| A | 7% | 14.2% | 13.1% |
| B | 12% | 21.4% | 17.8% |
| C | 9% | 32.3% | 35.1% |
| D | 12% | 24.2% | 22.4% |
Source: CashFlow Engine backtest data, October 2023 to 9 October 2026. Hypothetical results.
The Hunter's drawdown rules are enforced on the reference window the portfolio was fitted to. They are rules about the past. A portfolio that satisfied a 7% cap there then experienced 13-14% in the period that followed. The Hunter does not validate a forward edge, and these gaps are a measurement of how large the difference was in this sample.
How much of the in-sample result survived out of sample?
Roughly a quarter to a third. Walk-forward efficiency, the annualized out-of-sample profit divided by the in-sample reference profit, ran between 0.17 and 0.40.
| Setting | Monthly | Weekly |
|---|---|---|
| A (52-week reference) | 0.28 | 0.40 |
| C (26-week reference) | 0.17 | 0.25 |
| D (26-week reference) | 0.25 | 0.32 |
Setting B is not shown; no walk-forward efficiency was computed for it. A ratio of 1.0 would mean the out-of-sample period earned at the same annualized rate the portfolio showed on its fitting window. None of the settings came close. That gap is the difference between a result fitted to a past window and what the same portfolio did next in this sample.
Did the length of the reference period matter?
Within the rules of setting A, using total history earned about 17% more than a 52-week reference, with a worst drawdown about 62% deeper. That is one comparison on one rule set.
| Reference period | Profit | Worst drawdown |
|---|---|---|
| 52 weeks | +$52,812 | $11,304 |
| Total history | +$61,687 | $18,288 |
Source: CashFlow Engine backtest data, January 2025 to October 2026, reproducible search mode. Hypothetical results.
A 26-week reference was run only with the clock-limited search. It showed lower profit and a higher drawdown than the 52-week run, but those runs are noisy, as the next section explains.
What limits what these numbers can say?
Four settings, about three years of data, heavily overlapping comparisons and revisable backtest data. The numbers are descriptive of this sample and nothing beyond it.
- Four settings, about three years. One setting covers only about 21 months. Different settings, universes or periods could behave differently.
- Overlapping start dates. The head-to-head counts are not independent samples.
- Clock-limited searches are noisy. Two clock-limited runs of the total-history setting returned about +$26k and +$46k, against +$62k from the reproducible run. With the 52-week setting, the two modes agreed within 2%.
- Backtest data gets revised. Recent days in the library change after the fact. One dataset was replaced on 10 October 2026, and re-running the same as-of test later can shift the newest months by up to about $2.7k.
- Selection bias. The four settings were ones we already had saved. They were not drawn at random from every possible setting.
- No transaction costs on switching. See the turnover section above.
Frequently Asked Questions
- Should I re-run the Hunter every month?
- This article does not recommend a cadence. In our hypothetical backtest of four of our own settings, monthly re-picking produced more profit and a smaller worst drawdown than holding one selection, but out-of-sample drawdowns ran about 1.5-4 times the cap and results from four settings over about three years do not establish a forward edge.
- Is a Hunter portfolio still valid after a few months?
- The Hunter optimizes on the past window it was given and does not validate a forward edge. In the backtest, held selections earned less and drew down more than re-picked ones, which is consistent with a fitted selection going stale; that reading is an interpretation, not a proven cause.
- How much of the backtest profit survives out of sample?
- In this sample, roughly a quarter to a third. Walk-forward efficiency, annualized out-of-sample profit divided by in-sample reference profit, ranged from 0.17 to 0.40 across the settings shown.
- Does a drawdown cap in the Hunter limit real drawdown?
- It limits drawdown on the reference window the portfolio is fitted to. Out of sample, worst drawdowns in our backtest ran about 1.5-4 times the cap.
- How much does the portfolio change when it is re-picked?
- A weekly re-pick swapped 65-78% of the strategies each week and a monthly re-pick about 90% each month, so a re-picked portfolio was close to a new portfolio each time.
- Are these live results?
- No. All figures are hypothetical backtest results from the CashFlow Engine strategy library, run as of past dates without look-ahead. They do not represent actual trading.
Terms & Definitions
- Re-picking
- Running the same optimizer rules again on a later date using only data up to that date, then switching to the new selection.
- In-sample
- The past window on which a portfolio was selected; results there are fitted, not tested.
- Out-of-sample
- Data after the selection date that the selection could not have seen.
- Walk-forward efficiency
- Annualized out-of-sample profit divided by the in-sample reference-period profit.
- Maximum drawdown
- The largest peak-to-trough decline in the specified equity curve.
- Turnover
- The share of selected strategies replaced when a portfolio is re-picked.
Related reading
Educational disclosure: This article describes a hypothetical backtest of the operator's own Hunter configurations as a demonstration of the tool. It is not investment advice, not a recommendation of any setting, cadence or portfolio, and not a trading signal. The Hunter optimizes its rules on a past window and does not validate a forward edge. Backtested results do not reflect actual trading, past results do not indicate future results, and the figures are subject to selection bias. Options trading involves substantial risk, including the loss of the entire investment.