Rules by type
Every rule in the Rulebook, grouped by its job in a trade. Each line names the trader and links to the sources.
FilterWhat to trade and when to stay out
- System 1 skip rule: ignore a 20-day breakout if the previous breakout in that market would have been a winner. A loser is one that moved 2N against the position before a 10-day exit. If a signal is skipped, still enter at the 55-day breakout as a fail-safe.
- Crossover system: long-only on the S&P continuous contract. It is either long or flat, never short.
- Support/resistance system on gold, long and short. N-day support is the lowest low of the last N days, resistance the highest high. Long-term lines define the trend, short-term lines drive trading.
- Trend is up when price trades above long-term resistance and stays up until price trades below long-term support.
- From his tests on gold: trend lines under 100 days performed poorly (about 140 worked best), trading lines under 15 days performed poorly (30–40 worked best); the best pair was 120/30.
- Breakout scan: stocks in the top 1–2% by performance over 1, 3 and 6 months, with a prior move of 30–100% or more in the last 1–3 months.
- Then an orderly pullback with higher lows and a tightening range, price “surfing” the rising 10- and 20-day averages (sometimes the 50-day), lasting from about 2 weeks to 2 months.
- Episodic pivot: a gap up of 10% or more on big volume, ideally trading the average daily volume within the first 15–30 minutes, driven by news that catches the market off guard (earnings, guidance, FDA, contracts).
- Best on stocks that have gone sideways for 3–6 months or more; repeat pivots on stocks that already moved have a higher failure rate. For earnings pivots he wants strong sales and EPS growth and a large beat.
- Parabolic short: large caps up 50–100% or more in days or weeks, small caps 300–1000% or more, with 3–5 or more up days in a row.
- Daily charts are used for entry signals, not weekly or monthly ones; the weekly chart is for screening.“daily charts must be used for trade entry signals”
- Flags and pennants lasting more than four or five weeks are treated with suspicion; so is a corrective phase longer than a few weeks.
- For stocks, strict Edwards & Magee rules call for a pattern completion of at least 3% of the stock’s value.
- Says he sees about 10–20 really good patterns a year; most of the time the answer is to wait.
- Holy Grail (long): the 14-period ADX must first be above 30 and rising.
- Do not put capital to work in the middle of a trading range: buy new highs or pullbacks.“don’t put your capital to work in the middle of a trading range”
- For ordinary investors the same idea: get out of anything that falls below its 200-day moving average.
- Always be with whatever the predominant trend is.
- He will not buy anything that does not have both a great chart and good fundamentals; the approach is multi-disciplinary and not limited to equities (bonds, currencies).
- Trade only stocks in an uptrend; he will not trade against a trend.“I will not trade against a trend.”
- Prefer stocks with relative strength versus the overall market and use a “leadership profile” that compares a stock with common traits of past big winners.
- Hedging with put options only adds risk in his view (you can lose on both sides).
- Market condition: a confirmed uptrend in the S&P 500 is the main green light, and pullbacks in it are buying opportunities. He is cautious on individual small and mid caps when breadth is narrow.
- Work with the trend of the market: he says chances of success are three or four times what they would be if you buck the trend.
- Trade active stocks and do not tie capital up in sluggish ones.“In trading, keep on moving stocks!”
- A trader should be ready and willing to sell short as readily as buy for a long position.
- Trade about 100 futures markets across all asset classes; he says that universe is unrealistic for a small portfolio, though the idea works on a smaller number of contracts.
- Simple, purely mechanical logic; the same model and parameters for every asset class; price-based only.
- The bands give a relative sense of high and low: prices near the upper band are relatively high, near the lower band relatively low. Compare price action with indicator action inside that context.
- A touch of either band is only an observation, not by itself a trade signal.
- In strong trends price can ride along the upper or lower band for extended periods.
- The 20-period average and two standard deviations are starting points; settings may differ by market and purpose. The middle band should reflect the intermediate trend and not be tuned to give the best crossovers.
- When the average is lengthened, widen the deviation multiplier, and narrow it when shortening: about 2.1 at 50 periods and 1.9 at 10.
- Combine indicators that are fairly independent: a momentum gauge plus a volume gauge adds more than two momentum gauges.
- Trade a broad set of futures: his dataset has 146 instruments, and markets that fail cost or liquidity requirements are excluded. Instrument weights by asset class include 0.13 for metals and crypto.
- Drop any rule for an instrument whose trading costs exceed the “speed limit” of 0.01 Sharpe-ratio units; his actual costs are about 1.06% a year.
- Pullbacks: the market must be both trending and fluctuating. Skip very strong trends with no pullbacks; judge the trend from structure, swing length, lower-time-frame momentum and strength versus exhaustion.
- Avoid entering after potentially climactic large-range bars far outside the band, and trades that point to the end of a trend; do not force trades when the market is not pulling back.
- Scan for a close at least 4% above the previous close, on volume above the previous day’s and at least 100,000 shares (the formula in his shorthand: c/c1 ≥ 1.04 and v > v1 and v ≥ 100000). It works best on small and mid caps and lower-priced stocks.
- The stock should not be up three days in a row before the breakout; the day before should be negative or a range under 2%; the close should be within 20% of the day’s range from the high.
- Prefer an orderly pullback or base of 5 to 40 days with no 4% breakdowns inside it, a linear first leg if momentum is established, and avoid stocks whose past action was choppy.
- Universe: stocks that are members of the S&P 500. Trend filter: the close is above its 200-day moving average. Market filter: the 126-day return of the S&P 500 is above zero.
- Internal Bar Strength IBS = (close − low) ÷ (high − low) × 100, so a close at the low is zero. The idea: a low IBS suggests the next day closes higher.
- Reported in an all-trades test: keeping only trades with IBS under 10 (34% of trades) raised the average profit per trade by 58%; IBS under 25 kept 63% of trades with a 21% increase. Winners fell from 71% to 57% as IBS rose from 0 to 100.
- Three movements run at the same time: the primary movement (a bull or bear market, tending to last at least a year and generally much longer), the secondary movement (sharp rallies in a bear market and sharp reactions in a bull market) and the daily fluctuations.
- The two averages (industrials and railroads) must confirm each other: there is never a primary movement, and rarely a secondary one, in which they do not agree. They need not move on the same day or week; it is enough that they take the same direction.
- A “line” is a succession of closing prices over a sufficient number of days with fair trading volume within a narrow range (for example, industrials between 70 and 74 for weeks). He reads it as a period of accumulation or distribution.
- Use closing prices only. A break out of the line “will confidently indicate” a change in market direction of at least a secondary character, provided it is confirmed by the other average.
- For a single active stock the book quotes the same idea: if it stays in a narrow range (say two points) for long enough to form a horizontal line, that suggests accumulation or distribution, and a double top followed by another decline can lead to a further fall.
- High-of-Day (HOD) break: he prefers stocks above about $100 because their Level 2 is simpler to read, with large asks at half-dollar and dollar levels that serve as resistance and places to take partial profits.
- Candidates: stocks with high volume that gapped up or down, or have a strong catalyst (“stocks in play”). Use VWAP to judge whether a stock is too extended for a good risk. Do not enter in the direction of a long wick; take pullback entries only at clear support or resistance.
- Trend filter: the S&P 500 is above its 200-day simple moving average (SPY or E-mini futures work as proxies).
- Why 2 periods: he writes there is “little statistical evidence” for the standard 14-period RSI, and that the shorter period measures the short-term overbought/oversold state better.
- The only parameter is the look-back period. He tests 2 to 18 months; the best Sharpe ratios cluster at 12 months, also within every decade from the 1970s to 2012, so 12 months is the benchmark.
- Trade with intraday sentiment, measured by the NYSE TICK and Market Delta.
- Instrument: QQQ, which the paper calls the most liquid instrument that replicates the Nasdaq index; the leveraged variant uses TQQQ (3x the daily move of QQQ). Both long and short are traded.
- Test assumptions: 1 January 2016 – 17 February 2023, start capital $25,000, commission $0.0005 per share, no slippage in fills, data from Interactive Brokers.
- Sensitivity check: a stop of 5% of the 14-day ATR with an end-of-day exit gave a far higher result on TQQQ (9,350%). The authors themselves call this possibly unrealistic: the stop would be about $0.08 wide, which a larger account would likely overshoot.
- N = 1 is the traditional version. Test period 1 January 2000 – 30 June 2023, instruments SPY and QQQ.
- He personally uses N of 3 to 5 as a balance between fewer whipsaws and entering a new trend too late, and also uses the rule as a market-regime filter for other strategies.
- Trend filter: the close is above the 200-day moving average.
- For stocks from the S&P 500 and Nasdaq-100 he added: the index is above its 200-day average and the stock is above its own; 10 positions of 10% at most. Returns were only slightly above buy-and-hold with drawdowns about one-third smaller; 2013 – 2014 were strong mean-reversion years and 2015 was not.
- Test 2007 – 2014 (the start of the leveraged ETFs’ data). Three ways: rules found on QQQ applied to QLD, rules found on QLD itself, and QQQ signals used to trade QLD.
- Why a 2x fund is not twice as good: daily 2x tracking misses on big days (on 30 Sep 2008 QQQ rose 2.9% and QLD 10.4%), and the daily compounding works against holding it for long. In 2007 – 2013 QQQ and QLD were within 0.02% of an exact 2x on about 80% of days.
- Evaluate on the last trading day of each calendar month and trade at the next open. Four conditions: A — VIX at or below 25; B — the S&P 500 above its 200-day average; C — VWO (emerging markets) with positive 1-3-6-12 week momentum; D — BND (US bonds) with positive 1-3-6-12 week momentum.
- The momentum score averages the 1-month return × 12, the 3-month × 4, the 6-month × 2 and the 12-month return. Test period 1 January 2010 – 31 December 2023.
- Data: QQQ 5-minute bars from Alpaca, regular session only, January 2017 – September 2026. Development period 2017–2022 (1,504 sessions); out-of-sample 2023 – September 2026, kept for the final evaluation.
- Parameter search: 120 configurations (range 5–60 minutes, target 1R–4R, signal cutoff 10:30–13:00). The best on the development data was a 10-minute range, 3R, cutoff 10:30 (the “primary” variant), but it did not clearly beat the baseline.
- Overnight return = 09:30 open ÷ previous 16:00 close − 1; its size z = that return ÷ the standard deviation of the overnight return over the previous 60 days. Condition 1: z ≤ −2.0 (an unusually large drop).
- Condition 2: the price at 09:35 is above the 09:30 open (the first five minutes were up). Skip half-days and the 11 days before quarterly contract expiry; roll windows are excluded.
- The rule was chosen from 9 candidates (3 drop thresholds × 3 first-five-minute filters) using the training data only, and frozen on 14 October 2026. Data: NQ front month, 1- and 5-minute bars, June 2016 – October 2026, not back-adjusted.
- Robustness (out of sample): median +52.4 points; without the best trade +41.4; without the three best +26.4; bootstrap 95% range of the average +11.6 to +90.7 points.
- A trading day is the 24 hours from the close on day t−1 to the close on day t, split into: overnight, the first half-hour after the open, the middle of the day, the second-to-last half-hour and the last half-hour. “Rest of day” is everything except the last half-hour. For NQ the hours are 09:30–16:00 and the sample is 12 April 1996 – 1 May 2020.
- For comparison in the same table (rest of day): E-mini S&P 500 — coefficient 6.18, t = 4.97, R² 3.41%, out-of-sample 2.29%; Dow futures — 5.02, t = 4.12, R² 2.20%, out of sample 2.18%; all markets pooled — out-of-sample R² 2.88%.
- Data: CME NQ from Databento. Design period July 2024 – December 2025 (353 sessions) was used to choose rules; the holdout January – September 2026 (178 sessions) was checked once. Results include commissions and slippage, but the page does not state the amounts.
- Rejected on the holdout: absorption reversal at round numbers, VWAP and POC on tick data (−0.02R design, −0.29R holdout); value-area rotation, the “80% rule” (median +0.03R design, −0.10R holdout); failed auction at the previous high or low (−0.11R design, −0.06R holdout).
EntryThe trigger that opens a position
- System 1: enter when price exceeds by one tick the high (long) or low (short) of the preceding 20 days.
- System 2: the same with the preceding 55 days. Every System 2 breakout is taken, winner or not.
- Trade at the breakout price during the day; do not wait for the close or the next open. If the market opens through the breakout price, enter on the open.
- Add one unit every ½ N after the previous fill, up to four units in one market. The interval is counted from the actual fill price, so slippage moves the next level.
- Buy when the fast exponential average crosses above the slow one. Tested pairs: 15/150 as the base case, 85/325 as comparison; the page notes 90/330 as “pretty good values”.
- In an up trend, buy on a stop just above short-term resistance; in a down trend, sell short on a stop just below short-term support. Orders are set after the close for the next day.
- Buy the break of the opening-range high: the high of the first 1-, 5- or 60-minute candle. A daily-chart entry at the start of the breakout also works; buying in anticipation is, in his words, “not as effective”.
- Wait for the first 1-minute candle and buy the break of the opening-range high; add on 5-minute highs. If the range is missed, the 5- or 60-minute highs are used: “no need to be first in”.
- Short options: the opening-range low (1 or 5 minutes), the first red 5-minute candle, or a failed bounce into VWAP, the only intraday indicator he says he uses. “Wait” for the right setup.
- Prefers a breakout confirmed by a wide-bodied bar close, but may enter intraday when the breakout boundary is well defined.
- Does not chase: buying much past the initial breakout is “a losing practice”.
- A protective stop is attached to every entry order placed with an intraday stop. His initial entry is, as a rule, his final entry (no pyramiding); a bigger position at the breakout is taken for the same risk.
- Wait for a retracement to the 20-period exponential average (usually with the ADX turning down). When price touches it, place a buy stop above the high of the previous bar. Sells are the mirror image.
- Three-Day Unfilled Gap Reversal (buy): today the security gaps lower and does not fill the gap. For the next three sessions keep a buy stop one tick above the high of the gap-down day.
- Wolfe Wave (buy): wave 2 is a top, wave 3 the bottom of the first decline after it, wave 1 the bottom before wave 2 (point 3 lower than point 1), wave 4 the top of wave 3 (higher than the wave 1 bottom). A line from point 1 through 3, extended, projects the entry point (wave 5).
- She tries to trade in the direction of the most recent greatest swing on her time frame, and brackets a longer-time-frame entry (part at a bid, a buy stop higher in case of no fill).
- The day after an NR7 day she usually brackets stops around the early-morning range; in her tests it did not matter whether the first 45 minutes or the first hour was used.
- His signature setup is the Volatility Contraction Pattern (VCP). The interview does not state an exact pivot or breakout price rule.
- Example (earnings breakout): first entry of two tiers on the break above the earnings pre-market high, stop below the opening low.
- Second entry of two tiers on the break above the 200-day moving average on volume, with the stop below the significant pullback low of the opening consolidation.
- If the stock consolidates at the target and breaks to a new high, add two tiers again, with the stop below that consolidation low; later targets sell about half, then a quarter, of what remains.
- Worked example (U.S. Steel): when the stock looked ready for a sharp move, buy (or sell) 300 shares with a three-point protective stop.
- Add another hundred shares for every two points the stock moves in your favour, protecting each new lot with its own three-point stop; stop adding after roughly a ten-point move (about 800 shares).
- Go long if the current close is above the close 250 trading days (about a year) ago; go short if it is below. The system is always long or short (stop and reverse).
- Buy rule: buy when the monthly price is above the 10-month simple moving average.
- The model is updated once a month, on the last day of the month, with entries and exits at that close; price moves during the month are ignored.
- A close outside the bands is first read as a sign the move will continue, not reverse; this idea underlies many volatility-breakout systems.
- %b (where price sits relative to the bands) is used to spot divergences and patterns and to code systems; BandWidth identifies “The Squeeze” and can help signal trend changes.
- Use many rules, not one: momentum (EWMAC with 4, 8, 16, 32, 64 lookbacks), breakout (10 to 320 days), relative momentum, carry, skew and a little mean reversion. Trend-type rules get 0.6 of the forecast weight, the rest 0.4.
- Breakout rule: take the rolling max and min over the lookback; raw forecast = 40 × (price − midpoint) ÷ (max − min), then smooth it with an exponential average of span lookback ÷ 4. The 40 maps the natural range to roughly ±20.
- Use bands (Keltner or Bollinger) set to contain roughly 80–90% of price action and then do not change them. Wait for a bar to touch the band and enter near the average in the trend direction; a lower-time-frame break of the previous bar’s high is a useful refinement.
- Failure test: price trades beyond a prior pivot or support/resistance level and quickly fails back across it. Same-bar version: it closes back inside on the same bar, enter on that bar’s close. Next-bar version: a bar closes beyond the level and the following bar reverses back across it.
- Run the scan from the market open and enter as soon as a good setup appears, to catch the rest of the day’s move; alternatively scan at the end of the day and enter the next session.
- Entry: the 2-period RSI crosses below 2.5; buy on the next open.
- When both averages rise above the high of the line, the indication is strongly bullish: in a bear market it can mean a secondary rally, and in 1921 it meant the start of a primary bull movement.
- Do not use the HOD break in the first 10 minutes of the session. Entry options: below the HOD (early, or on surging volume, but no further than $0.10 below), at the HOD only when it is an exact dollar level with a large ask being bought up, and in most other cases just above the HOD after the bid breaks it.
- 5-minute opening range breakout: start 5 minutes after the open; the high and low of the first 5-minute candle define the range. Enter in the direction of the break (it may come a few minutes after the range is set), or on a pullback to a technical level such as VWAP.
- Day 1: the 2-period RSI is below 65 (neutral to possibly oversold). Day 2: the 2-period RSI closes lower than on day 1. Day 3: it closes lower than on day 2.
- Buy at the close of day 3.
- Absolute momentum is positive when the asset’s excess return (its return less the Treasury-bill return) over the look-back period is positive; then hold the asset.
- Reevaluate once a month. For stock indices and similar assets he does not skip the latest month in the look-back, because such assets are less subject to liquidity and microstructure issues than single stocks.
- Enter on a pullback in sentiment, not at the extremes: do not chase price highs or lows. Wait for a higher low in an uptrend, a lower high in a downtrend, or a hold of support or resistance in a range.“Don’t chase price highs or lows.”
- Direction is the direction of the first 5-minute candle: if it closed up, buy at the open of the second 5-minute candle; if it closed down, sell short at that open. No position if the first candle is a doji (open = close).
- Buy at the next open after the close has been above the 200-day moving average for N days in a row.
- Original rule: buy at the close when the close is a 7-day low of closes. Version used for the portfolio test: buy at the next open with 100% of cash.
- Buy at the close when the 2-period RSI is below 5 and the price has fallen for two or more days (the buy comes on the second down day). No trend filter.
- All four true: 50% UPRO (3x S&P 500) and 50% TQQQ (3x Nasdaq-100).
- One or two false: 50% QQQ and 50% SPY. Three or four false: 100% TLT (long Treasury bonds).
- Baseline rule: opening range 09:30–09:45; from 09:45 to 12:00 buy if a candle closes above the range high, sell short if it closes below the low; entry at the next candle’s open; at most one trade per session.
- Buy 1 NQ contract at 09:35 New York time.
- The measured signal: the return of the rest of the day has the same sign as the return of the last half-hour on average. In the paper’s regression for NQ the coefficient on rest-of-day return is 6.36 (t-statistic 7.97), in-sample R² 4.10% and out-of-sample R² 3.76%, as printed in the table.
- Kept: “initial balance acceptance breakout” — design +0.19R (232 trades), holdout +0.19R (117 trades). Kept: 15-minute opening range breakout — design +0.14R (312 trades), holdout +0.14R (64 trades).
StopWhere the idea is wrong
- Initial stop 2N from the entry (long: below, short: above). One N is 1% of equity, so the maximum risk of a trade is 2%.
- When a unit is added, raise the stops of earlier units by ½ N, so the whole position sits 2N from the latest unit.
- Protective stop for a long goes below short-term support, for a short just above short-term resistance. The position is closed by that stop; no separate profit target.
- Stop at the low of the day, and not wider than the stock’s ADR or ATR (a stock with 5% ADR means a stop of at most 5%).
- Stop at the low of the day, within 1× (at most 1.5×) the average daily range or ATR.
- Parabolic short stop: the high of the day, or a reclaim of VWAP. Target: the 10- and 20-day averages, where these stocks usually bounce; risk/reward about 5–10×.
- If a fill comes better than the stop, he reads it as a warning and exits immediately; trading errors are liquidated at market at once.
- After the fill, place a protective sell stop at the newly formed swing low and trail it as profits grow.
- Gap reversal: if filled, the protective sell stop goes at the low of the gap-down day; accrued profits are protected with a trailing stop; if not filled within three sessions, cancel the order.
- Get out at single-digit losses, which requires using a stop-loss; the only way to manage risk is to sell for a small loss before it becomes a large one.
- When adding, move the stop on the whole position below the new consolidation low: the closer stop keeps total risk near the planned percentage.
- Limit risk almost always: make it a habit to place a two- or three-point stop behind any trade made to profit from fluctuations. He notes that Harriman thought three-eighths of a point to one point was enough, but Harriman started as a floor trader.
- No stops: a position is reversed only when the opposite signal appears.
- Place the stop beyond the previous extreme; as a rough starting point, 2–4 ATRs beyond the entry.
- Failure test stop: at or just beyond the new extreme set on the move (for a short, just above the high of the signal bar).
- Stop at the low of the entry day: the trade only makes sense if there is immediate follow-through.
- Place the initial stop where pullbacks rarely reach, usually just beyond a support or resistance zone or a key swing point (often “a few pennies” beyond a high or low). When that point is far from the current price, use a volatility stop instead.
- Study of 3,068 breakouts upward from ascending triangles, double bottoms and rectangles (US stocks, 1991–2008): price fell below the breakout price about 80% of the time, by 5.3% on average (median 4.2%), and the lowest low came about 11 days after the breakout on average.
- A stop 2 cents below the breakout-day low was hit within a month 61.5% of the time with an average loss of 2.19%; a stop below the lowest low of the previous 10 days was hit 14.9% of the time with an average loss of 7.53%.
- Share of samples that did not drop beyond each level in the first month: 48% at 4%, 75% at 8%, 89% at 12%. Tight stops under 4% were hit about half the time.
- Examples by pattern: double bottom, a penny below the lower bottom; cup with handle, a nickel or so below the handle’s lowest valley; rectangle, a few pennies below the bottom for an upward breakout; flags and pennants, at the bottom of the pattern.
- For an ascending broadening wedge broken upward, a stop 67% of the way down the latest swing low-to-high held about two-thirds of the time, he reports.
- Opening-range stop: the loss or break of VWAP (a tight stop, since you should not be far from VWAP). If entered on a pullback at VWAP, the stop is the next technical level or slightly beyond VWAP.
- Put the stop far enough away that random noise does not trigger it, so that being stopped out clearly means the trade is wrong.
- Stop at the low of the first 5-minute candle for a long and at its high for a short; the distance from entry to stop is the risk, $R.
- Stop on the opposite side of the opening range. If a stop and a target are hit in the same bar, the stop is assumed to come first.
- Because of NQ’s one-minute noise the author reports structure stops of 50–80 points, which means 1–4 MNQ contracts on a $2,000–3,000 drawdown limit.
ExitHow profit is taken or a trade is closed
- System 1 exit: a 10-day low (long) or 10-day high (short), all units at once. System 2 exit: 20 days.
- Exit when the fast average crosses below the slow one. No separate stop-loss is described. Signals come after the close, orders fill at the next open.
- Sell one third to one half after 3–5 days and move the stop to breakeven. Trail the rest with the 10- or 20-day average depending on speed; beginners use the 10-day and the first close below it.
- Usually takes profits at targets. Exits trades that are underwater on Friday and does not take a losing trade home over the weekend.
- If the pattern does nothing after touching the average (no gain, no loss for several days), she says there is no reason to be in the trade: “don’t give it the benefit of the doubt”.
- Wolfe Wave target is the line from point 1 to point 4; the initial stop goes just beneath the new reversal at point 5 and can quickly be moved to breakeven. One cannot start looking before points 1–4 exist.
- Most often she gets out when the market fails to do what she expected; the best patterns give instant feedback. When a trade feels uncomfortable she may place an exit-on-close order to remove the stress of timing the exit.
- On scalps she does not trail stops and exits in one piece: the shorter the time frame, the higher the noise and the more likely a trailing stop is hit.
- His main metric is the 200-day moving average of closing prices: if a market trades below it, he gets out. He says he applied it to stay flat at the top of the 1987 crash.“You play defense, and you get out.”
- He says he has never used a stop-loss, “not once”; he sells when the reason he bought has changed. A falling price does not by itself mean sell: it means re-examine the thesis.“I’ve never used the stop loss. Not once.”
- Hold only while the potential reward exceeds the risk and while the stock is “behaving”; protect profit once the trade works. He describes himself as a swing trader.
- At the first target (about 50 cents above the earnings high): if the stock turns down, sell about 75% and reassess; if it keeps going, sell only about 25% and trail the stop below the target.
- Reasons to sell a working position: the price target is reached, the intraday trend breaks, or breaking market news.
- Take profit on a further advance or raise the stop so that at least a solid profit is locked in.
- He quotes a speculator he knew who threw a position out at once if it fell two or three points after he bought, on the ground that his first judgement was wrong.
- Sell rule: sell and move to cash when the monthly price is below the 10-month simple moving average.
- Let winners run: do not exit winners without reason. Admit errors quickly and exit; use time stops when the expected scenario does not develop; being flat is a legitimate position.“Do not exit winners without reason!”
- Take the first profit at a gain equal to the initial risk (1R), then scale out of the rest. Add a new pullback trade only after taking partial profits and moving stops, so you never hold more than one trade’s risk.
- The target is a 3–5 day momentum burst of roughly 8% to 20%. If follow-through is weak, consider exiting by day 3.
- Once the stock is 4–5% above the entry, move the stop to breakeven intraday. After 3 to 10 days of consecutive gains sell into strength, preferably intraday, and keep a small portion with a tight stop (25% in his example).
- A trailing stop raised as price rises and never lowered gave an average rise of 2.42% (look-back 0 days) up to 5.66% (10 days); a volatility trailing stop matched the 5-day look-back at 3.73%.
- Exit: the 2-period RSI closes above 50; sell on the next open.
- When the averages break below the low of the line, the market has reached a “saturation point”: it signals a secondary bear movement in a bull market or the start of a primary downward movement.
- If an HOD break is not working 10 minutes after the open, exit within about a minute; in the afternoon wait at most about 5 minutes. Small losses (around 0.2R) or break-even exits are normal; the full 1R “fail-safe stop” happens when the HOD is violently rejected within three 1-minute candles, about 10% of trades.
- Take profits in aggressive 50% partials: the first near 1R (about a $1 move on these stocks), then 50% of the rest at each further move of about $1, or $2 for more volatile stocks.
- Target: the next technical level (moving averages, the previous day’s close, the high or low of the last two days, or a marked daily support/resistance). Do not stop yourself out early when a winner reverses: take enough partials.
- Sell when the 2-period RSI closes above 75.
- When the excess return turns negative, exit the asset and move into 90-day Treasury bills until absolute momentum is positive again.
- Set profit targets first, from range extremes, pivot prices and the R1–R3 / S1–S3 levels, and decide where the market is likely to move before entering.
- Size so that you can take partial profits at the first target and still hold a piece for further movement; add if the trade proves strong; take quick profits if conditions look shaky at the first target.
- Target 10R, or close the position at the market close if 10R was not reached, whichever comes first. No partial profit-taking.
- Sell at the next open after the close has been below the 200-day moving average for the same N days in a row.
- Original rule: sell at the close when the close is a 7-day high of closes. Portfolio version: sell at the next open.
- Sell at the close when the 2-period RSI is above 50.
- Target 2R; any open position is closed at 15:00.
- Sell at 10:00. No stop and no target.
Size and riskHow much to put at risk
- Measure volatility as N: the 20-day exponential average of True Range (what is now called ATR).
- One unit = 1% of the account divided by (N × dollars per point), so a move of one N on one unit equals 1% of equity. Volatile markets get fewer contracts.
- Position limits: 4 units in one market, 6 in closely correlated markets, 10 in loosely correlated ones, 12 in one direction in total.
- Risk per lot = 5 × ATR. Risk budget (“heat”) = 10% of current equity. Position = budget ÷ risk per lot, rounded to a whole contract.
- Size = equity budget ÷ risk per lot, where risk per lot is the distance between the entry stop and the protective stop.
- No more than 30% of the account in one stock overnight; most positions 10–20%; risk typically 0.25–1% of the account per trade.
- Risks about 60 basis points (0.6%) per trade.“I risk 60 or so basis points on each trade.”
- A 60% win rate system still produces five losers in a row at some point out of ~200 trades about 90% of the time, so a wide-stop system with such odds needs minuscule leverage or many markets (about 20).
- Aim for five-to-one: risk one dollar to make five. At that ratio a 20% hit rate is enough, so “I can be wrong 80% of the time”.
- Make large concentrated bets where you have conviction; a big position keeps your attention and, he argues, lowers risk compared with 35–40 names.
- He recalls that Soros thought the 1992 pound trade was good enough that he should have been at 200% of capital: sizing should follow the quality of the idea.
- “Rule of risk” from The PlayBook: on every A+ setup, put 30 percent of the intraday stop loss at risk.
- After each profitable month, raise the intraday stop (and with it the size on A+ setups) gradually.
- Think in percentages of the daily stop, not in dollars or shares: one “tier” = 10% of the daily stop; the example plan risked 40% in total.
- Use only a small part of capital for trading, not over five or ten per cent of his “loose capital”, so that no unexpected event can cripple him.
- Size every position by volatility so that each has the same average daily impact on the portfolio (ATR-style). Volatility is estimated from the past month; he admits that assumption does not hold exactly.
- Rebalance regularly to keep the risk level constant as volatility changes.
- The tested portfolio spreads equally over five asset classes (US stocks, foreign stocks, bonds, real estate, commodities), each held only while above its average.
- Results exclude taxes, commissions and slippage. He notes turnover of the timing system near 70% versus about 20% for buy-and-hold, which matters a lot for taxes.
- Scale forecasts to an average absolute value of about 10; values near 20 are rare. Positions follow the forecast, so a stronger signal means a larger position.
- Volatility estimate: 70% recent volatility and 30% the average over the last 10 years; momentum-type forecasts are reduced when volatility is above its historic level.
- The instrument diversification multiplier is estimated and capped at 2.5; correlation between instrument sub-systems averages about 0.05.
- Dynamic optimisation finds the integer-contract portfolio with the lowest tracking error to the ideal one, and buffering reduces trading costs.
- Watch correlations: several positions can amount to one oversized risk. Manage today’s correlations, not the past ones. Risk management is the most important part of the job.
- Trade failure tests with smaller size and risk than most other setups: gaps through the stop and larger-than-expected losses happen, and in trends the gap-and-go beyond an extreme tends not to reverse.
- For long-term compounders the Kelly criterion maximizes the expected logarithm of wealth: bet a fixed fraction of current capital chosen by edge and variance, recomputed as capital changes.
- Overbetting is much more severely penalized than underbetting, so edge estimates should be conservative.
- “Half Kelly” has 3/4 of the growth rate with much less chance of a big loss: the chance of ever losing half the starting capital is 1/2 at full Kelly but only 1/8 at half Kelly.
- Doubling before halving has probability 2/3 at full Kelly and 8/9 at half Kelly.
- Why fractional Kelly: if the true edge is half the estimated one and you bet 1.5 times the estimated Kelly fraction, the growth rate is negative and you are ruined; even full Kelly then gives a growth rate of zero with wild swings.
- Risk the same dollar amount on every trade (“R”). A trade that stops out fully loses 1R; he does not hold for a fixed profit or a full 1R loss, because the setup should work quickly.
- Costs: 20 basis points are deducted for each switch into or out of Treasury bills; the number of switches ranges from about 0.33 a year (REITs) to 1.08 a year (high-yield bonds).
- Keep size small enough to absorb a stop-out without strain.
- Require at least 1:1 reward-to-risk to the first target and meaningfully more for later targets.
- Risk 1% of the account per trade: shares = int(min(A × 0.01 ÷ $R, 4 × A ÷ P)), where A is the account, P the entry price, and 4 is the maximum intraday leverage. The 1% is chosen because the average daily move of QQQ is about 1%.
- The 4x leverage cap binds: the authors estimate that 60% of trades were taken with an exposure at least 40% below the one needed for 1% risk. TQQQ is used so the same dollar moves fit within the cap.
DisciplineBehaviour rules around the trade
- Take every signal. The document says most of a year’s profit may come from two or three big trades, and the Turtles who were dropped were those who failed to enter when the rules said so.
- Decide the exit point before entering. A stop is a non-negotiable exit: when price trades there, the position is closed, every time.
- The document calls the exits the hardest part: waiting for a 10- or 20-day extreme can mean watching 20–40% or more of the open profit disappear.
- His “Trading Tribe Essentials”: ride your winners, cut your losses, manage your risk, use stops, stick to the system, file the news.
- After two consecutive signals fail immediately he watches for a trend reversal. Win rate hovers around 50%; the horizon is one to three weeks; he tracks closed-trade equity, not unrealized profit; a 5% drawdown is about as much as he wants to endure.
- Losing streaks come from sloppy habits, trader error or stubbornness; a small loss should never become a big one. Switching methods at their drawdown low leaves you one step out of sync.
- The core question is how to keep from losing everything; he has seen too many things go to zero, stocks and commodities.
- Keep ego out of it; do not “get crazy” when a position is going up; do not confuse a genius with a bull market. He admits that after selling everything in January 2000 he bought back in and lost about $3 billion because emotion overrode discipline.
- Have in advance an answer to: when to stop trading after making money, which market conditions suit your system, what you do in a losing streak, and what you do when you break your own rules.
- Either make a business of trading or do not try to be a trader: training, not an occasional dab.
- His reported result before fees: about 22% a year, 26% maximum drawdown, Sharpe ratio 1.1 (a backtest he presents as a demonstration).
- He states the rules “are of course not perfect” and the system cannot be traded with a small account: it is a learning tool, not a complete strategy.
- Reported on the S&P 500 from 1901 to 2012: invested about 70% of the time, fewer than one round trip a year, the worst drawdown cut from 83.66% to 42.24%, yet it lagged buy-and-hold in roughly half of all years.
- He says a trend-following model can underperform buy-and-hold in a roaring bull market; the benefit is lower volatility and drawdown, not a higher average return (average returns were 11.26% vs 11.22%).
- Do not assume a normal distribution: with default settings about 90% of data falls inside the bands in practice.
- The bands do not give continuous guidance; they help identify setups where the odds may favour you. They work on any instrument and bar length with enough trading activity.
- Profitable trading requires an edge: if you cannot name your edge, you do not have one. Markets are highly competitive and nearly fully efficient, and simple approaches tend to work.
- People who watch a large downward swing tend to quit or be forced out; he reports most prefer the comfort of half Kelly in exchange for giving up a quarter of the growth.
- Conclusion of the paper: long-term compounders should avoid betting more than the Kelly fraction and, since probabilities are uncertain, should limit their investment fraction enough to prevent a significant risk of overbetting.
- He concludes that IBS alone did not work as a stand-alone mean-reversion signal: it only looks at one-day mean reversion. The test takes all signals; it is not a portfolio test.
- The theory is a barometer of the market’s direction, not a guide to individual trades: it says nothing of stops or position size, and the averages are read as a whole, not by single stocks.
- He says 95% of his trades are HOD breaks: strategy focus forces patience and fights overtrading.
- He reports a simulation from 1995 to 2006: 102 trades, 84.31% “correct”, 1,013.9 S&P points gained, an average holding period of 5.76 days.
- He uses the 12-month benchmark “to minimize transaction costs and the risk of data snooping”, and applies the rule to eight assets from 1974 to 2012, then as an overlay on a 60/40 and a risk-parity portfolio.
- Do not second-guess trades during the session; review them after the close.
- Reported on QQQ: 1,795 trades (51% long, 49% short), win rate 24%, average 0.13R per trade, Sharpe 1.12, total return 675% against 169% for buy-and-hold QQQ, max drawdown 22%, annualized alpha 33% (p = 0.0025).
- Reported on TQQQ with the same rules: total return 1,484%, 0.18R per trade, Sharpe 1.19, max drawdown 28%, alpha 48%. Buy-and-hold TQQQ in the same period: 438% with an 82% drawdown.
- The authors name the main risk as operational, not the strategy itself: slippage, turning a day trade into a swing or buy-and-hold position, and exiting too early before the target.
- On SPY, per his text: N = 1 sharply cuts the maximum drawdown but lowers the annual return by about 1.2 points against buy-and-hold; N = 2 nearly halves the number of trades and trails buy-and-hold by only 0.12 points; N = 3 to 6 cut drawdown a lot for a small loss of return.
- On QQQ, per his text: for every N from 2 to 10 the return is higher, the maximum drawdown lower and the number of trades smaller than in the table’s baseline rows.
- SPY and QQQ, 2000 – 2007: it beat buy-and-hold with about 26% time in the market and a clearly smaller drawdown, although the return figures were modest.
- SPY and QQQ, 2008 – 2015: both returns were below buy-and-hold; SPY’s drawdown was much lower, but QQQ’s was about as bad as before. He traced the underperformance to a single bad trade in 2008.
- His conclusion: the strategy may be too simple, may have worked while mean reversion was less crowded, and may now have little or no edge.
- For QQQ, a 2x result would be a return of about 21.6 with a maximum drawdown near 22; applying the rules directly to QLD raised the return by about 50% but the drawdown by more than 150%, so QLD did not deliver a clean doubling. Trading QLD on QQQ signals came nearly to the expected doubling.
- He says he spent about five minutes on the rules, so he expects no strong results; the post is about method, not a strategy to trade.
- Reported with the original settings: an annual return of 24.4 and a maximum drawdown of 54; the year 2022 returned −48%, and 2023 returned +64%.
- His doubts: overfitting is easy and the number of combinations tried and the test dates are unknown; an emerging-markets filter for US ETFs looks odd; picking TQQQ is an obvious choice with hindsight; he asks whether in-sample and out-of-sample tests were done; a 54% drawdown would be hard to live with; defaulting to TLT worked in the long bond bull market but hurt in 2022.
- Development period, baseline, before costs: 1,472 trades, +0.056R average, profit factor 1.13, maximum drawdown 17.6R.
- Out of sample, before costs: primary variant 906 trades, win rate 42.6%, +0.074R average, +67.3R in total, profit factor 1.15; baseline 928 trades, +0.070R. By year (primary, total R): 2023 +20.8, 2024 +18.9, 2025 +35.3, 2026 to September −7.7; the other variants were also negative in 2026.
- With hypothetical costs of 4 cents per share round trip the primary variant’s average falls to +0.054R (48.4R in total); at 10 cents to +0.022R (20.2R). Under the 4-cent case the bootstrap 95% interval for the average net R includes zero.
- The author states the work is not a validated live-trading system and that positive historical expectancy is necessary but not sufficient. He also notes that this is not a replication: the rules differ from the paper (close-confirmed breakout, next-bar entry, different range, target and cutoff).
- Net of costs (1 tick slippage per side plus $4.50 fees, about 0.7 points round trip; 1 point = $20): train 2016-08…2021 — 38 trades, +7.9 points, t = 1.86; validation 2022–2023 — 8 trades, +49.3, t = 1.54; test 2024 – October 2026 — 10 trades, +52.2, t = 1.87.
- All out-of-sample trades together: 18 trades, +50.9 points net on average, 61% winners, t = 2.49 (p ≈ 0.01). The worst trade lost 120 points (25 January 2022).
- Limitations listed by the author: only 18 out-of-sample trades (about 4 a year); the rule was picked from 9 candidates and later periods were looked at more than once; the effect is not monotonic (drops of −2 to −1.5 do not rebound); many trades cluster on stress days where real slippage may be worse; costs are modelled from trade prints, not quotes; no risk management, one contract each.
- Next step announced by the author: forward paper trading from 14 October 2026 with no parameter changes, aiming at 20–30 new trades.
- The effect reverts over the following days, and the authors link it to the hedging of short-gamma positions (market makers in options and leveraged ETFs buy when prices rise and sell when they fall, so they push the move at the close).
- Parameter rankings did not carry over between the periods (rank correlation −0.27), so the author kept the simple version of each setup instead of optimizing.
- The author says the edge is unproven until a forward test of 30 or more journaled trades, and that the project is educational, not advice.
Rules are our paraphrase of the sources; only short quotes are verbatim. This is information, not advice or a promise of profit.