Double 7s (Connors/Alvarez via pinkfish, Farrell Aultman) - SPY Daily Close-at-7-Day-Low Buy / Close-at-7-Day-High Sell, SMA200-or-SMA70 Regime Filter, 15% Stop, 2x Reg-T - Faithful Port on USEQ

Faithful translation of the pinkfish 'double-7s' example strategy by Farrell Aultman (MIT licence, https://github.com/fja05680/pinkfish/blob/18a0d557ead25e0e82de3ece86bcab6c325a59ed/examples/strategies/double-7s/strategy.py). The strategy comes from Larry Connors & Cesar Alvarez, 'Short Term Strategies that Work'. Credit goes to the author and the book. Instrument: SPY.USEQ (the source's own symbol), DAILY bars, long-only, one position at a time. The venue keeps the source's economics: the ETF it was designed for, commission-free US equity execution (~0.02-0.05% RT spread+impact modelled) and Reg-T 2x margin, which matches the source's margin=2. This makes it an exception to lesson 167: the port does not need to change its cost regime, because USEQ IS the regime it was written for. USEQ is 2.5% of the corpus and the mechanism is short-horizon ETF mean reversion, which is not a crypto re-mine. PARAMETERS, the values in the source's main() and strategy.ipynb (the configuration the author actually runs): period=7, sma=70, stop_loss_pct=0.15, margin=2 (leverage 2.0), use_regime_filter=True, use_adj=False. The module-level default_options differ: stop_loss_pct=1.0 (the stop sits at price 0, so it is effectively no stop), margin=1, sma=200, and everything else is the same. Pre-register the default_options set as a SECONDARY variant scored separately. Do not optimise between the two. INDICATORS, all on the daily CLOSE series only (the source drops open/high/low): (1) regime = pinkfish CROSSOVER(fast=1, slow=200). It is positive while close > SMA(close,200), so regime>0 is equivalent to close > SMA200. (2) sma = SMA(close, sma=70). (3) period_high = rolling max of CLOSE over the last 7 bars INCLUDING today. (4) period_low = rolling min of CLOSE over the last 7 bars INCLUDING today. Warm-up: no trading until all indicators are defined, i.e. at least 200 bars (the source uses dropna=True). Data note: our USEQ prices are split-adjusted and dividends are not in PnL. The source uses use_adj=False on raw closes, which also excludes dividends, so the PnL basis is consistent. Expected flat state: in the market roughly 25-35% of days, flat the rest of the time.

Do not change the strategy rules (period=7, sma=70, regime_sma=200, stop 15%, leverage 2). Make two execution fixes. (1) Cap entry notional below buying power: e.g. capital_fraction 0.90, or size from free margin net of impact. Do not re-submit an entry while an entry order is in flight. (2) Re-run and confirm two things: entry_diag.dropped_entry_inflight falls from 136 to about 0, and trades appear in 2001-02..04 and 1998-09..11 when the rules signal. Metric expected to move: annualized_volatility should fall from 51% toward about 20%, with the paired -X/+X flat-day returns gone. MTM Sharpe should then rise from 0.15 toward the realized-consistent level. Economics are fine: avg_trade 0.69% against a 0.05% USEQ floor, PF 1.56. If the flat-day paired swings persist after the sizing fix, flag the run for engine review rather than optimizing. ## Library refinements (from the knowledge library; test them, do not assume them) The library has no Double 7s source, but it holds Connors/Alvarez RSI2 material from the same book (200-day MA regime, close-based entries, Connors' finding that stops hurt, and an exit on a close below SMA200), plus lumibot guidance on sizing against buying power. It supports the analyst's execution fix: size below buying power and net of the open position. The rule variants belong in separately scored, pre-registered secondaries, not in the primary. 1. [sizing] Size entries against remaining buying power, not full equity x leverage: Keep all signal rules. Change position_size so that qty = floor(min(equity * 0.90 * leverage, free_buying_power * 0.98) / close) - current_open_qty. Here free_buying_power = equity * leverage - open notional, and the 0.98 factor leaves room for USEQ spread and impact. Return 0 if an entry order is already pending or working (check self.cache.orders_open / orders_inflight for the instrument). In practice: capital_fraction 0.97 -> 0.90, plus the in-flight guard. — The analyst found 136 dropped in-flight entries, signalled trades missing in 2001 and 1998, and paired -X/+X flat-day swings that push annualized vol to 51% and MTM Sharpe to 0.15 even though avg trade is 0.69% and PF is 1.56. The lumibot margin-aware example computes the maximum size from portfolio value times a leverage cap, then subtracts current position size to get 'available capacity' before sizing. It stresses that mark-to-market cash is the real buying power, so orders never exceed what the account can carry. At 0.97 x 2.0 the order sits almost exactly at the Reg-T limit, so any adverse move or impact between signal and fill makes it unaffordable. (source: Continuous futures - no expiration to manage (lumibot docs, 'Example: Tracking Available Buying Power') p.1) 2. [stop] Pre-registered secondary: no hard stop (stop_loss_pct = 1.0): Run a separately scored secondary with stop_loss_pct=1.0 (stop disabled) and everything else unchanged: period=7, sma=70, regime_sma=200, leverage 2.0, capital_fraction 0.90. This is the source's own default_options stop setting. Compare stop-exit trade count and PnL with the primary. Do not optimise the stop level. — The RSI2 write-up of Connors' method (Short-Term Trading Strategies That Work) says 'Connors does not advocate the use of stop loss… the use of stop-loss actually damaged performance' in tests of hundreds of thousands of trades. On an at-the-low mean-reversion entry, a fixed 15% stop mostly fires into capitulation lows, which are the best entries for this setup. The 35.7% max drawdown at 2x leverage should show whether the 15% stop actually caps tail loss or just locks it in. This is a claim to test, not proof. (source: RSI2 Mean Reversion Strategy using in futures (Larry Connors RSI2) p.1; Larry Connors拉里·康纳斯 RSI2均值回归策略 p.1) 3. [regime] Pre-registered secondary: exit on a close below SMA200 (regime exit): Run a secondary that adds one exit condition: while long, sell at the close if close < SMA(close,200) AND close < SMA(close,70), i.e. when the entry regime (the 'or' of the two MAs) no longer holds. Evaluate it on the close with no look-ahead. Entries, period=7, the 15% stop and leverage 2.0 stay unchanged. Score it separately from the faithful primary. — The MQL5 port of the Connors RSI2 Classic rules from the same book uses 'Exit buy when: last close price > 5-period moving average, or last close price < 200-period moving average'. The trend filter works as an exit as well as an entry gate. Double 7s currently holds a 2x-levered long through a regime break until either a 7-day closing high or a 15% loss. That is the likely source of the 35.7% drawdown in the 2001-02 and 2008-style declines. The article also warns about fat-tail risk and large floating drawdown for RSI2-type mean reversion. (source: Day Trading Larry Connors RSI2 Mean-Reversion Strategies p.1)