AB Adaptive Strength Methodology

@techreport{disalvatore2026abas,
  title        = {AB Adaptive Strength: A Regime-Governed, Relative-Strength Model Book with a Technically Confirmed Protection Basket},
  author       = {di Salvatore, S. and E1 (Emergent)},
  institution  = {Assets Bulletin Research},
  year         = {2026},
  month        = {September},
  type         = {Working Paper},
  number       = {v0.1},
  url          = {https://assetsbulletin.com/research/ab-adaptive-strength}
}

1Motivation — the missing allocation layer

The two earlier desk papers stop at the state of the market. The Triaxial Regime Hazard produces a posterior P* and a rung — the odds that an equity drawdown regime begins — and Suppressed Hazard Convergence dates whether that hazard is realizing. Neither says what a model book should hold, and the usual way of closing that gap — a discretionary committee reading the odds and choosing names — imports the two weaknesses the reads were built to remove: unstated rules and unmeasured turnover. From January to September 2026 the Elite model book was cut by a multi-model creator under a governor; it produced defensible names and an ungovernable process. What was needed was a book whose composition follows from the reads by rules that a reader can replay, so that every line on the page has a cause on the page.

The design principle is a division of labour. Adaptivity lives in the sleeve sizes, where the desk's reads are strongest and slowest-moving; selection is mechanical relative strength, where discretion is weakest and the literature is oldest (Levy 1967; Jegadeesh & Titman 1993). Protection is a basket that earns its weight rather than a fixed hedge that pays carry in every calm month (Faber 2007; Hurst, Ooi & Pedersen 2017). The result is a book that changes its risk with the regime and its names with the tape, and states both.

2Literature and the gap

Momentum selection, volatility management, trend-following hedges and regime-conditional allocation are each well studied; what is rarely done is to put them in one book with the regime read from an external observer rather than inferred from the book's own returns, and to publish the turnover, the modelled cost and the look-ahead beside the result. Ang & Bekaert (2004) show the optimal mix changes with the regime; Daniel & Moskowitz (2016) show where a strength book breaks; Gârleanu & Pedersen (2013) show why one should not trade all the way to the target. AB Adaptive Strength is an operational answer to the three at daily frequency with free data — in the spirit in which TRH made time-varying transition probabilities operational and SHC made the volatility paradox observable.

3The model

3.1 The Protection Sleeve Governor

Let rₜ be the TRH rung, φₜ the cycle phase and σₜ the SHC state at the close before a cut. The governor is a lookup, not a fit:

G(rₜ, φₜ, σₜ) = DEFENSIVE  if rₜ ∈ {HIGH, SEVERE} ∨ φₜ = Phase 3 ∨ σₜ ∈ {ACTIVE, CONVERGED, REALIZED}
             = GUARDED    if rₜ ∈ {MODERATELY ELEVATED, ELEVATED} ∨ φₜ = Phase 2
             = CALM       otherwise

The budgets are the whole of the adaptivity. A DEFENSIVE read does not veto any growth name; it cuts the growth budget from 85 to 50, quadruples the protection budget from 5 to 20 and — since d-161 (Sep 25, 2026) — swaps two of the ten growth slots for the XDTE + SOXY income pair, so a guarded or defensive book carries eight growth names and the income line while a calm book carries ten and none. This is deliberate: the reads are good at how much and poor at which, and a veto on names in a regime with a handful of observations is the kind of rule a backtest cannot support (section 7).

3.2 The growth sleeve — relative strength in a strict source order

Every candidate passes one screen on the last completed close: an intact uptrend, no MACD bearish cross, no close at the upper Bollinger band with expanding ATR, never a leveraged or inverse instrument — and, since d-156 (adopted Sep 25, 2026 on the third run, R6–R7), an RSI cap set by the tape read off SPY at the cut: on a strong tape (SPY above its 50-day and less than 4 % off its 52-week high) RSI below 85; on a weak tape RSI below 70 and at most 15 % above the 50-day, and for a high-beta name (252-day β ≥ 1.5) RSI below 65 and at most 12 %. Before d-156 the cap was RSI below 75 in every tape. The ablation tables of section 5 (R4–R10) were run on the screen as it stood at the time and stand as published; the reconstruction of R1–R3 is rebuilt after every close on the engine as built, so from Sep 25, 2026 it carries d-156 and d-158.

tape = weak  ⇔  P_SPY < SMA₅₀(SPY)  ∨  P_SPY / max₂₅₂(SPY) − 1 ≤ −4 %
screen(i) ⇔ Pᵢ > SMA₅₀(i) > SMA₂₀₀(i)  ∧  ¬macd_bear_x(i)  ∧  ¬bb_atr_stretch(i)  ∧  RSI₁₄(i) < cap(tape, βᵢ)
cap(strong, ·) = 85 · cap(weak, β < 1.5) = 70 with Pᵢ / SMA₅₀(i) − 1 ≤ 15 % · cap(weak, β ≥ 1.5) = 65 with ≤ 12 %        (d-156)
rank by  R₁₂₆(i) = Pᵢ,ₜ / Pᵢ,ₜ₋₁₂₆ − 1        (six-month total return)
rotation in (d-154):  RSᵢ = Pᵢ / P_SPY · ratio = 100·RSᵢ / SMA₆₃(RSᵢ) · momentum = 100·ratio / SMA₂₁(ratio)
                       improving ⇔ ratio < 100 ∧ momentum ≥ 100 ∧ momentum − momentum₍ₜ₋₅₎ > 0

The slots — ten in a CALM read, eight in GUARDED or DEFENSIVE (d-161) — are filled tier by tier, best six-month return first inside each tier, one name per theme (cybersecurity, semis ETF, memory, semis equipment, cloud/data, AI software, crypto proxy, EV, broad tech ETF, banks, payments, medical devices; since d-158 one per region — Japan, China, broad EM, developed ex-US, India, Latin America, Korea, Europe) so the sleeve cannot become one trade under ten tickers. Each name weighs the growth budget divided by the number of names, rounded to 0.5 % (85 / 10 = 8.5 %; 65 / 8 = 8.125 → 8.0 %; 50 / 8 = 6.25 → 6.5 %) — equal weight, following DeMiguel, Garlappi & Uppal (2009) on the robustness of 1/N — and the rounding residual, like an unfilled slot, rests in the bills line rather than being forced into a weaker name. A name already held stays while it ranks inside the hold band of the slot count plus two (≤ 12 in CALM, ≤ 10 otherwise): this hysteresis is the book's only turnover control on the growth side, and the ablation in section 5 is what set its width. Should the book fall below twelve lines — few protection names confirmed, a thin screen — the growth sleeve floor-fills from the next names through the screen, then from the next uptrend names that fail only the RSI cap, each fill flagged on the page.

Rotation in as a trigger (d-154). Six-month strength is slow by construction: a name that lagged for a year and has begun to lead the index will not rank inside the top ten for months. The desk's Rotation card reads that turn with the JdK relative-rotation graph — relative strength against SPY smoothed over 63 sessions, its momentum over 21 — and the sleeve reserves two of its ten slots for names in the improving quadrant with rising momentum, computed on the same closes so the cut and the card agree. They must still pass the screen (an intact uptrend is required — a laggard below its 200-day is not a turn, it is a hope) and sit outside the protection basket, which keeps its hedging role. A rotation-in name stays while it remains improving or leading; when it turns weakening it is judged like any other name by the six-month rank. The pool itself widened in the same amendment: the financial and health-care mega caps the earlier bi-weekly book carried (LLY, UNH, ISRG, ABT, JPM, GS, BAC, V, MA, SPGI) are candidates beside the tech names, so a sector rotation out of tech has somewhere to go inside the rules.

3.3 The protection basket — weight only on confirmed momentum

The basket holds 53 assets that have hedged equity drawdowns in different macro regimes — metals, currencies, bitcoin, duration, TIPS, linear inverse, long volatility, managed futures, commodities, oil, international markets, defensive sectors, low-volatility factors and defensive mega-caps. None of them hedges every regime, and each of them costs carry when it is not hedging (Baur & Lucey 2010 on gold; Kaminski 2011 on managed futures). The rule therefore asks each name, on the close before the cut, whether it is hedging now:

confirmed(j, t) ⇔ all five checks hold at t  ∧  at t − 1      (two consecutive closes to enter)
held name leaves ⇔ a check fails at t  ∧  at t − 1              (two consecutive failures to leave)
weight = 4 % a slot · best 3-month return first · one name per theme AND per macro cluster · Σ ≤ protection budget

The two-close rule is the anti-whipsaw clause of the desk doctrine (d-030, d-151): protection names are the sleeve most prone to passing a screen on one close and failing it on the next, and a hedge that is entered and exited within a fortnight is a cost with no protection. The cluster rule came from the first live shadow, which confirmed BNO, PDBC and XLE together — three tickers, one oil trade. Caps: a linear inverse line at most 8 % and only one, long volatility (VIXY) only when SHC has recorded a touch (d-122). Whatever budget is not confirmed rests in the bills line: an unconfirmed hedge is not replaced by a weaker one.

3.4 The bills line and the income pair

The bills line is the book's cash: it carries its own budget, every unfilled growth slot, every unconfirmed protection slot and the 0.5 % rounding residual of the growth sleeve (either sign). Its vehicle is chosen by the d-121 capture rule for the hold window — WEEK (weekly T-bills) for a fortnight window, unless SGOV or BIL carry a yield edge of at least 25 bps or an ex-date inside the window. The income pair, XDTE (weekly index premium) and SOXY (monthly semiconductor income), enters only in a GUARDED or DEFENSIVE read, where it takes the two growth slots the regime gives up (d-161), half the income budget each; it exists so that part of a cautious book is paid to wait, in the rebalancing cadence of the book itself. In a CALM read there is no income line and its budget sits in growth.

3.5 Cadence and execution

The book is decided at the close before the first session on or after the 2nd and the 16th and executed at that session's open (doctrine d-064). Positions are held in exact fractional shares to four decimals, so a 20 % sleeve reads 20.00 % rather than the 19.94 % of whole-share lumps and no cash line appears. Bounds (d-161): never fewer than 12 and never more than 16 lines — 10 growth + bills + up to 5 protection slots in CALM; 8 growth + XDTE + SOXY + bills + up to 5 protection slots otherwise — no leveraged instrument, one broad inverse at most. An expected cut is disclosed at least 36 hours ahead (d-141).

4Data and method

Daily adjusted closes, opens and dividends of the growth pool (89 names), the protection basket, the income pair, the bills vehicles, SPY and QQQ from January 2024 (Alpha Vantage first, Yahoo as fallback). The reconstruction starts with $100,000 of cash on Jan 2, 2025, cuts at the close before every 2nd and 16th and executes at the next open; dividends are reinvested in the paying name; SPY total return is computed on the same basis (the Tactical Tech convention, d-146). Three proxies stand in for reads that did not exist before launch: the regime is read off the tape (SPY below its 200-day or ≥ 8 % off its high → DEFENSIVE; ≥ 4 % off, QQQ RSI > 70 or below the 50-day → GUARDED), the M8 accumulation tier is a rules proxy (above the 50-day, one-month return beating SPY, 20-day on-balance volume rising), and an SHC touch is never assumed. The pool is the desk's pool as it stands today — a look-ahead by construction, since every name in it is a name that mattered enough to be in the pool in 2026. The line is gross of costs; a drag of 5 bps a side on every traded dollar is modelled and reported beside it, never netted into it. Every threshold is stated; nothing is fitted in-sample beyond the two ablation decisions of section 5, which are disclosed as such.

5Results

6Reading the results

The excess return is the pool's, the shape is the governor's. A relative-strength book on a 2026 pool over a 2025–26 tape led by semiconductors will beat the index; that is what look-ahead does and the paper says so. What the design can claim is the shape: a beta near 0.8 and a correlation near 0.6 to SPY for a book that is 50–80 % in strength names, obtained by shrinking the growth budget when the tape proxy turned. Table R2 also shows the price of that: in the sessions the proxy read DEFENSIVE — March to May 2025, the tariff drawdown and its rebound — the book lagged SPY, because a strength book de-risked into a V-shaped recovery is the momentum-crash setting of Daniel & Moskowitz (2016). The governor is not free; it is a premium paid for the drawdowns that did not rebound.

Two “free” refinements, two costly rules kept. The tighter hold band and the 4 % protection slots each add return and remove drawdown on this sample and cost nothing in governance, so they were adopted. The one-name-per-theme rule and the M8 priority tier each cost return on this sample and were kept anyway: the first is a concentration rule, the second is the desk's own read entering first, and neither was ever an alpha claim. A methodology that dropped its governance rules whenever a backtest said so would not be a methodology.

The defensive branch is the least tested part of the book. The tape proxy produced a handful of DEFENSIVE cuts in twenty-one months; the sleeve design exists for exactly those cuts. The live governor, reading the actual rung, phase and SHC state from launch, will produce the observations the proxy cannot, and the paper's tables will carry them as they arrive.

7Disclosure, not rule — the beta question

The desk doctrine carries a beta veto (d-051, d-068): in later phases, names with a 252-day beta above 1.5 leave. Applying it to the AB growth sleeve in GUARDED and DEFENSIVE regimes cut the reconstruction's return from +52 % to +28 % without improving the drawdown (table R4) — the classic momentum result that the strength book's beta is time-varying and highest exactly when the names are strongest (Grundy & Martin 2001). Applying it in DEFENSIVE only changed nothing, because there were five such cuts. A rule that either destroys the return or has no observations is not a rule the paper can defend; a number the reader should see is. The book therefore publishes, at every cut, the equal-weight 252-day beta of the growth names and the names a beta ≥ 1.5 veto would have removed — and applies nothing. Should the live record show the sleeve's beta doing damage in a DEFENSIVE regime, that disclosure is the evidence the rule would need.

8Relationship to TRH and SHC

TRH gives the odds; SHC says whether the suppression that keeps those odds from realizing is exhausted, and dates the moment; AB Adaptive Strength says what a rules book holds given both. The direction of dependence is one way — the reads govern the book, the book never feeds the reads — so the pairing is not circular and the book's record becomes a third validation surface: every cut carries the rung, the phase and the SHC state that sized it, and the sleeve returns by regime (table R2) are the reads' own scorecard in allocation space. A DEFENSIVE cut that protected is evidence for the rung; one that cost is evidence against it. That is what a governed book is for.

9Limitations

One sample of twenty-one months, one pool with look-ahead, a tape proxy for the regime and a rules proxy for the M8 read; seven ablation variants in the first run, six in the second and thirty-one in the third (R4–R9) on that one sample, which is the situation Bailey & López de Prado (2014) warn about, so no Sharpe on this page is an estimate. The third run's tape-state split (R7) rests on twelve weak-tape cuts, five of them in a V-shaped recovery — a reading, not a law. The protection checks are technical proxies for mechanisms — safe-haven flows, trend, carry — that are only observable with positioning data; the two-close rule buys stability with a session of lag. Costs are modelled at a flat 5 bps a side; the true cost of a 35 % one-way turnover every fortnight in a small book is lower, and in a large one higher. Every number on this page is a statistical observation published by a financial publisher — not a forecast, not a track record, not advice.

10References

1. Ang, A. & Bekaert, G. (2004). How Regimes Affect Asset Allocation. Financial Analysts Journal, 60(2), 86–99.
2. Asness, C. S., Moskowitz, T. J. & Pedersen, L. H. (2013). Value and Momentum Everywhere. Journal of Finance, 68(3), 929–985.
3. Bailey, D. H., Borwein, J. M., López de Prado, M. & Zhu, Q. J. (2014). Pseudo-Mathematics and Financial Charlatanism: The Effects of Backtest Overfitting on Out-of-Sample Performance. Notices of the American Mathematical Society, 61(5), 458–471.
4. Bailey, D. H. & López de Prado, M. (2014). The Deflated Sharpe Ratio: Correcting for Selection Bias, Backtest Overfitting, and Non-Normality. Journal of Portfolio Management, 40(5), 94–107.
5. Barroso, P. & Santa-Clara, P. (2015). Momentum Has Its Moments. Journal of Financial Economics, 116(1), 111–120.
6. Baur, D. G. & Lucey, B. M. (2010). Is Gold a Hedge or a Safe Haven? An Analysis of Stocks, Bonds and Gold. Financial Review, 45(2), 217–229.
7. Cooper, M. J., Gutierrez, R. C. & Hameed, A. (2004). Market States and Momentum. Journal of Finance, 59(3), 1345–1365.
8. Brav, A. & Lehavy, R. (2003). An Empirical Analysis of Analysts’ Target Prices: Short-term Informativeness and Long-term Dynamics. Journal of Finance, 58(5), 1933–1967.
9. Daniel, K. & Moskowitz, T. J. (2016). Momentum Crashes. Journal of Financial Economics, 122(2), 221–247.
10. DeMiguel, V., Garlappi, L. & Uppal, R. (2009). Optimal Versus Naive Diversification: How Inefficient Is the 1/N Portfolio Strategy? Review of Financial Studies, 22(5), 1915–1953.
11. Faber, M. T. (2007). A Quantitative Approach to Tactical Asset Allocation. Journal of Wealth Management, 9(4), 69–79.
12. Frazzini, A., Israel, R. & Moskowitz, T. J. (2014). Trading Costs of Asset Pricing Anomalies. Fama-Miller Working Paper; Chicago Booth Research Paper No. 14-05.
13. Gârleanu, N. & Pedersen, L. H. (2013). Dynamic Trading with Predictable Returns and Transaction Costs. Journal of Finance, 68(6), 2309–2340.
14. George, T. J. & Hwang, C.-Y. (2004). The 52-Week High and Momentum Investing. Journal of Finance, 59(5), 2145–2176.
15. Grundy, B. D. & Martin, J. S. (2001). Understanding the Nature of the Risks and the Source of the Rewards to Momentum Investing. Review of Financial Studies, 14(1), 29–78.
16. Hamilton, J. D. (1989). A New Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle. Econometrica, 57(2), 357–384.
17. Harvey, C. R. & Liu, Y. (2015). Backtesting. Journal of Portfolio Management, 42(1), 13–28.
18. Hurst, B., Ooi, Y. H. & Pedersen, L. H. (2017). A Century of Evidence on Trend-Following Investing. Journal of Portfolio Management, 44(1), 15–29.
19. Jegadeesh, N. & Titman, S. (1993). Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency. Journal of Finance, 48(1), 65–91.
20. Jegadeesh, N., Kim, J., Krische, S. D. & Lee, C. M. C. (2004). Analyzing the Analysts: When Do Recommendations Add Value? Journal of Finance, 59(3), 1083–1124.
21. Kaminski, K. M. (2011). In Search of Crisis Alpha: A Short Guide to Investing in Managed Futures. CME Group.
22. Korajczyk, R. A. & Sadka, R. (2004). Are Momentum Profits Robust to Trading Costs? Journal of Finance, 59(3), 1039–1082.
23. Levy, R. A. (1967). Relative Strength as a Criterion for Investment Selection. Journal of Finance, 22(4), 595–610.
24. Lo, A. W. (2004). The Adaptive Markets Hypothesis: Market Efficiency from an Evolutionary Perspective. Journal of Portfolio Management, 30(5), 15–29.
25. Moreira, A. & Muir, T. (2017). Volatility-Managed Portfolios. Journal of Finance, 72(4), 1611–1644.
26. Moskowitz, T. J., Ooi, Y. H. & Pedersen, L. H. (2012). Time Series Momentum. Journal of Financial Economics, 104(2), 228–250.
27. Novy-Marx, R. (2012). Is Momentum Really Momentum? Journal of Financial Economics, 103(3), 429–453.
28. Wilder, J. W. (1978). New Concepts in Technical Trading Systems. Trend Research.
29. di Salvatore, S. & E1 (Emergent) (2026). Triaxial Regime Hazard (TRH): A Decomposed Time-Varying Transition-Probability Framework for Equity Drawdown Risk. Assets Bulletin Research Working Paper v1.1.
30. di Salvatore, S. & E1 (Emergent) (2026). Suppressed Hazard Convergence (SHC): Dating the Realization of an Elevated Equity Drawdown Hazard from Suppression Capacity and Price Pressure. Assets Bulletin Research Working Paper v0.1.