Core-Satellite Portfolio With Tactical Allocation

Every great strategy has a tradeoff. The aggressive ones (HAA Leveraged, leveraged trend, accelerated dual momentum with capital-efficient ballast) get you 15 to 23% CAGR over the 1992 to 2026 window tested below, but spend a year or more under water at their worst. The defensive static portfolios (Cockroach, Permanent, Quartet) cap drawdowns at 20 to 30% but their CAGR caps too. Pick one and you live with whichever pain you signed up for.

The interesting question is what happens if you stop picking. Hold five strategies with genuinely different mechanisms and let the diversification do the work.

Here is the answer over a 33 year backtest using only strategies already in our catalog: 13.82% CAGR, -20.5% maximum drawdown, Sharpe 0.93. The Sharpe beats the best of the five components held alone, HAA Leveraged 2x (0.90), at a little over half its drawdown. The drawdown is lower than the best of them. That is not a coincidence, that is the point.

(A note on language up front: most of this 33 year window is technically in-sample for at least some of the components. HAA was published in 2023 and ADM in 2018, so calling the full window "out of sample" would be sloppy. We will use "backtest" throughout. The forward-looking caveats are in their own section below.)

What "core-satellite" actually means

Core-satellite is a portfolio architecture, not a marketing label. It splits your capital into two roles:

• Core = the part of the portfolio that does not need to be right about regime, trend, or momentum. It is built to deliver acceptable returns through any market environment by holding genuinely different asset classes (or capital-efficient overlays of them) and rebalancing back to fixed weights. Static. Boring on purpose. The drawdowns are bounded by the diversification, not by any signal that might fail.
• Satellites = the parts that try to add return when conditions are right, and step out of the way when they are not. Tactical. Each satellite carries its own signal logic, its own definition of "the trend just changed", its own way of going defensive. They earn higher CAGR than the core, but they also spend more time wrong.

The reason to combine them rather than pick one is that they fail differently. The core does not have a bad year because of a missed signal, it has a bad year because every asset class went down. The satellites have bad years when their signal whipsaws. Those failure modes do not overlap, so when you put them in the same portfolio, the bad weeks of one are usually offset by the calm weeks of the others.

The traditional core-satellite split (in financial advisor land) is roughly 70-80% core / 20-30% satellites. We will use 60% core / 40% satellites because the satellites here are themselves diversified TAA strategies, not single-stock concentrated bets. They earn their larger weight by being conservative tactical rather than aggressive single-name.

The five strategies

Two static cores plus three tactical satellites, weighted 60/40 with even splits inside each tier:

The deliberate design choice: the two cores share zero rules with the three satellites, and the three satellites each respond to a different signal. HAA reads the absolute momentum of TIP, its canary. TQQQ Trend reads SPY's 200-day trend and QQQ vs SPY relative strength. ADM reads accelerated relative momentum between US and international stocks, falling back to Treasuries, with the SmartStack overlay providing capital efficiency. So when one satellite gets a regime call wrong, it is rare for the other two to be wrong on the same axis at the same time.

A note on ADM's lineage: the dual-momentum framework comes from Gary Antonacci's 2014 book and his GEM strategy. ADM (Accelerating Dual Momentum) is a separate variant developed by EngineeredPortfolio (the blog from Chris Ludlow), which averages the 1, 3, and 6 month relative momentum and applies it to US large caps (SPY) vs international small caps (SCZ), with TLT or TIP as the safe asset. The "SmartStack (Gold+MF)" wrapper is our overlay: it holds 3x funds at one third of the weight and puts the freed capital in gold (GLD) and managed futures (KMLM). So ADM SmartStack inherits Antonacci's dual-momentum chassis, EngineeredPortfolio's acceleration on the signal, and our ballast on the defensive leg. Three layers, three different authors, all credited.

Why core/satellite specifically (and not just five-strategy equal weight)

You could argue that 20/20/20/20/20 across all five strategies would also blend their risks. It would, but you would lose two structural properties that matter:

1. The cores anchor the portfolio when all three satellites are wrong at once. 2008 had this character. Trend filters were late. Dual momentum signals oscillated. The static cores held bonds and gold the whole way through. Equal-weighting would have given the satellites 60% of the portfolio, three times their share of the diversification benefit. The 60/40 split keeps the anchor heavy enough to absorb the rare regime where every signal fails together.
2. The satellites earn their weight by being only moderately correlated with each other, not just with the cores. If we let a single satellite drift to 30% it would dominate the satellite tier and we would be back to "pick the best one and hope". Cap satellites at roughly equal weight inside their tier, and you get the satellite-of-satellites diversification too.

The 60/40 split is a judgment call, not a math optimum. We tested splits from 70/30 down to 40/60 and the Sharpe stays in a 0.90 to 0.96 band. Lower core means more CAGR, while the drawdown barely moves across the range (-20.4% to -21.2%). Calmar edges up as the core shrinks (0.63 at 70/30, 0.67 at 60/40, 0.73 at 40/60), so we keep 60/40 for the heavier anchor, not because the numbers peak there.

What the blend actually did, 1992-11-30 through 2026-05-01

This is the window of our first run in May 2026, when November 1992 was the latest common start of the five sleeves. The cards now go back further (see the window section below), and we kept these dates so the restated figures compare like for like with the originals. That is 33.4 years and roughly 8,400 trading days, with monthly rebalancing back to the target weights. The figures below were recomputed on 5 October 2026 from each card's current backtest run, again on 13 October 2026 after the Quartet's card was rerun, on 15 October 2026 after three cards were rerun on a new managed-futures history, on 16 October 2026 after HAA Leveraged 2x and ADM Stack were rerun on index data for early international and emerging-market prices, and on 17 October 2026 after every card was rerun with longer warm-up histories.

Read the bold column carefully. The blend's Sharpe of 0.93 beats all five components, the best of which is HAA Leveraged 2x at 0.90. Its Sortino of 1.33 is above HAA's 1.29, and its Calmar of 0.67 is above the best of the five, HAA Leveraged 2x's 0.65. And the maximum drawdown of -20.5% is shallower than the best individual sleeve (Cockroach at -21.8%), even though three of the five components have drawdowns deeper than -25%.

This is what diversification across genuinely different mechanisms gets you: you sacrifice some peak CAGR (over 1992 to 2026 the blend's 13.8% sits below HAA Leveraged's 23.5%, ADM Stack's 15.8% and TQQQ Trend's 15.5%) but you keep the bulk of the average and you cut the painful drawdowns roughly in half compared to the most aggressive component.

How it held up across regimes

A few honest observations:

• The 2000 to 2003 dot-com bust was the blend's strongest regime relative to the S&P, which lost nearly half its value peak to trough. The blend gained 8.6% annualized and the worst it ever felt was -14.6%. That is mostly the trend filters in HAA, TQQQ Trend, and ADM doing their job, while the static cores held gold and long bonds.
• The 2008 GFC was the hardest regime. Blend lost 1.2% across the 21 month window, with a -20.5% drawdown along the way, its deepest ever. Bonds and gold inside Cockroach, plus the Quartet's bond-stack legs, did most of the heavy lifting on the way down. The satellites caught the recovery on the way back.
• 2022 was the second-hardest regime. Stocks and long bonds fell together and gold was flat. None of the five sleeves made money on the year: HAA Leveraged 2x (-1.3%) and Cockroach (-3.1%), with its defensive sectors, gold and Treasuries, held up best, while the Quartet and ADM Stack each lost about 20%. Blend lost 11.6% on the calendar year, about two thirds of the S&P 500's 18% loss that year.
• COVID is interesting. The blend fell 20.2% in March 2020, in the most violent four-week sell-off in modern history, almost as far as in 2008. By December 2020 the blend was up roughly 27% on the year. The drawdown was real, the recovery was fast.

Calendar year statistics

• Positive calendar years: 30 out of 35 (86%).
• Worst calendar year: 2022 at -11.6%.
• Best calendar year: 1995 at +38.8%.
• Worst rolling 12 months: -18.9% ending October 2008.

That is a portfolio profile that very few investors associate with "five tactical and leveraged strategies inside it". The aggressive components are aggressive, and yet the blend behaves like a moderately aggressive but well-mannered portfolio.

Two real critiques worth addressing up front

This is a public methodology piece and the smarter readers in the TAA community will spot two design tensions immediately. Better to say them ourselves.

The leverage cost inside the blend

The Quartet runs roughly 1.55x effective leverage through NTSX, GDE, and RSST, which means that side of the portfolio is paying T-bill plus a financing spread to get the synthetic exposure. For futures-based funds like these three that spread sits inside the futures price and no filing discloses it, so treat any number as an estimate. Our backtests charge Fed funds plus 0.5% on the leveraged notional for the years before each of these funds launched. Swap-based funds cost more, about 2% to 4% of the fund a year in the filings we read, and since 5 October 2026 our backtests charge each of them its filed spread on the full swap notional instead. None of the five strategies here is affected. On the Quartet's 30% slice that is roughly 16% of the portfolio in leveraged notional, so the spread costs roughly 8 basis points a year at that rate. Cockroach holds no cash, so no part of the blend sits in T-bills while the Quartet pays to borrow.

The backtest window matters more than the headline implies

Our window starts in November 1992 because that was the latest common start of the five components when we first ran this in May 2026. The cards have since been extended back (HAA Leveraged to 1974, TQQQ Trend, ADM, Cockroach and the Quartet to January 1985), so the longest window all five now share starts in January 1985. On that window, to October 2026, the blend returned 14.82% a year with the same -20.5% maximum drawdown. We kept the 1992 window for the table so the restated figures compare like for like with the originals. 33 years sounds like a lot, and it is, but the period from 1992 onward inherits two regime tailwinds that are not normal across all market history.

Long bonds have been in a structural bull market for most of this window. The 30 year Treasury yielded around 7% in late 1992 and bottomed near 1% in 2020. That single regime did most of the heavy lifting for the Quartet's ZROZ leg and for Cockroach's Treasury sleeve. HAA's card now reaches back to 1974, but the Quartet and Cockroach still start in 1985, so the blend cannot be tested through the 1973 to 1981 inflation cycle, when the Quartet's long bond legs would have lost meaningful ground. Gold tells a similar story in reverse: post-1980 was a multi-decade bear that our window misses most of.

What this means in practice: take roughly 1.5 to 2 percentage points off the published CAGR before forecasting forward, and assume the maximum drawdown could be a few hundred basis points worse than -20% in a long-rate-rising regime. The Sharpe is more stable across regimes than the CAGR but it is also lower than 0.93 in unfavorable bond and gold environments. The headline numbers are a best-of-recent-history reference. The expected forward path is probably lower.

Why this works (the un-mysterious version)

There is no magic in the math. When you blend five return streams whose pairwise correlations average somewhere around 0.4 to 0.6, the volatility of the blend falls faster than the average of the volatilities. Sharpe is return-over-volatility, so dropping volatility while keeping a healthy portion of the return raises Sharpe. The same diversification effect compresses drawdowns: when one sleeve is in its worst drawdown, the others are usually not, so the portfolio-level drawdown is shallower than any individual member's.

What does require some thought is which strategies you put in. Five strategies that all read the same signal (say, five momentum strategies on different lookbacks) will not give you this result, because their drawdowns happen at the same time. The blend above works because the five mechanisms are genuinely independent:

• Cockroach uses no signal at all. Five static holdings: three defensive stock sectors, Treasuries and gold.
• Quartet uses no signal at all. Four static capital-efficient ETFs.
• HAA Leveraged reads a binary canary (TIP absolute momentum) to flip between offensive and defensive baskets, holding 2x leveraged ETFs.
• TQQQ Trend reads the 200-day moving average on SPY, plus QQQ vs SPY relative strength, to switch between leveraged Nasdaq with managed futures, RSST, and a defensive basket.
• ADM SmartStack reads accelerated dual momentum (1/3/6-month average) between US large caps and international small caps, with TLT or TIP as the safe asset and the SmartStack gold + managed futures overlay.

So the satellites can disagree with each other (and with the cores) without all rotating into the same trade.

A simpler implementation: RSST as a single-ETF core

If maintaining a 4-ETF Quartet feels like too many moving parts, there is a cleaner version where the second core collapses into a single ETF: RSST itself. RSST is Return Stacked U.S. Stocks & Managed Futures, a wrapper that gives you 100% S&P 500 exposure and 100% managed futures overlay on the same dollar. Two return drivers in one ticker.

Replace the Quartet's 4-ETF static with 15% RSST + 15% ZROZ as your second core. Now the portfolio is:

You drop the GDE gold leg and the NTSX bond+stocks leg from the second core. Cockroach already provides gold (20% of its 30% slice = 6% portfolio), so that leg is partially preserved. You lose a bit of capital efficiency on the equity side but the simplicity gain is real. On the same 33-year window, this simplified blend produces:

• Blend CAGR 13.42% (vs 13.82% original, a 40 bps cost)
• Max drawdown -19.3% (vs -20.5% original, slightly shallower because the synthetic Quartet pre-2018 carries a touch more equity drag than direct ZROZ)
• Calmar 0.69 (vs 0.67 original)

The risk-adjusted profile is essentially the same. The cost of simplification is a little under half a percent in CAGR. The benefit is one fewer rebalance to think about per year and three fewer ETF tickers in your account. Reasonable trade for a smaller portfolio or anyone who values low operational overhead.

You can push the simplification one step further and drop Cockroach too, going to a pure 2-core RSST + ZROZ design. That variant tested at CAGR 13.81% / MaxDD -21.3% / Calmar 0.65 with 30/30/13.3/13.3/13.3 weights, but the Sortino drops (1.27 against 1.33) and the drawdown deepens because you lose the gold leg from Cockroach. We would not recommend going this minimal unless the satellite tier already captures gold exposure (ADM SmartStack's overlay holds some gold). The Cockroach + RSST + ZROZ + 3 satellites version is the better simplified target.

What we are not telling you

A blog post that only shows the headline numbers is not honest. The caveats:

• All five sleeves use simulated prices before their funds launched. The Quartet uses a synthetic NAV chain for RSST before its September 2023 inception (built from the underlying components). TQQQ Trend uses a Nasdaq leverage proxy before TQQQ's 2010 launch (computed from the daily 3x reset on the underlying index, with realistic financing assumptions), and most of its defensive sleeve is rebuilt from managed-futures proxies before 2020, the years that give it most of its edge over simply holding T-bills. HAA Leveraged 2x holds 2x leveraged ETFs, simulated from their underlying indexes before those funds launched, and Cockroach's sector funds and ADM SmartStack's funds are chained to older proxies the same way. We disclose all of this on the strategy detail pages and in the backtest engine logs.
• The simpler variant uses synthetic RSST and ZROZ. RSST's real ETF only goes back to September 2023. ZROZ's real ETF goes back to 2009. Pre-inception we use BestFolio's price-builder fallback chains (S&P 500 plus a managed-futures index for RSST, long-duration Treasury yield curve construction for ZROZ). The chains are documented in fallback_table.py and visible in the price build log on every backtest.
• Implementation friction is real. Five strategies of varying mechanics means somewhere between 12 and 20 ETFs in actual execution at any given month. For accounts under $50K, the rebalancing tax and the cost of fractional shares chip noticeably into the published returns. The numbers above are gross of trading costs and assume no slippage.
• Past performance, etc. A 33 year backtest is a lot of ground but it is still one path. The forward path will look different.
• Tax inefficiency. The satellites turn over monthly. In a taxable account, that pulls realized short-term gains forward. The blend works best in a tax-deferred wrapper (IRA, 401k, ISA, PEA, depending on jurisdiction).
• You still have to follow the satellites every month. Cockroach rebalances quarterly and the Quartet annually, and both require almost no attention. The three satellites need a monthly trade. Skipping one because life happened that week is the most common failure mode of any tactical portfolio, and it does not get easier when you have three of them.

How to test this in BestFolio

Every component above is in the strategy catalog at bestfolio.app/strategies. The static cores are free to inspect. The two SmartStack and Leveraged variants are Pro. To replicate either blend yourself:

1. Open /portfolios in BestFolio.
2. Create a new portfolio and add the strategies above with the listed weights.
3. Hit "Run backtest" to see the same chart we ran for this post.
4. Use the live signals page to track the satellites monthly. Cockroach rebalances quarterly and the Quartet once a year.

The numbers in this post will reproduce, modulo a few basis points of difference if you tweak the start date or use slightly different rebalancing dates.

Bottom line

The honest reason to combine five strategies instead of picking one is not that any one of them is broken. HAA Leveraged at 23% CAGR over 1992 to 2026 (26.5% on its full card, from 1974) is not broken. The reason is that the maximum drawdown of the aggressive strategies runs from roughly 25 to 37%, and most investors will not hold through that. A 20% maximum drawdown is something a normal person can sit through. That alone justifies trading away some of the peak CAGR.

The bonus is that the Sharpe ratio does not go down: the blend's 0.93 beats the best single component's 0.90, with a drawdown about half as deep. You would expect to pay for the lower drawdown with a lower risk-adjusted return. Not having to, when the components are only moderately correlated, is one of the few free lunches finance hands you. Core-satellite is the architecture that makes the free lunch eatable: anchor with strategies that cannot be wrong about regime, accelerate with strategies that can. The math takes care of the rest.

All numbers in this post come from BestFolio's backtest engine on production data. They were first run on 2026-05-01 and restated on 5, 13, 15, 16 and 17 October 2026 from each card's current run, on the same window and weights. Since 23 September 2026 the engine also charges that financing for the years before each stacked fund launched, which is one reason the Quartet's figures are lower than in the first version. Strategy metrics are pulled directly from the prod database. The blend computation script is reproducible and uses monthly rebalancing back to the target weights. If you want the raw NAV series for the blend, ask in the comments.

Correction, 5 October 2026: the results in this post were restated on current production data, on the same window and weights. The first version, run on 1 May 2026, gave the blend 15.19% a year, a -19.2% maximum drawdown and a 1.07 Sharpe, and said that Sharpe beat all five components. Since then the strategy cards have been rerun with a longer history, corrected price data and, since 23 September 2026, a financing charge on stacked funds, and four of the five components now return less on this window. The blend is now 13.67%, -19.8% and 0.90, a Sharpe level with HAA Leveraged 2x rather than above it. The regime table, calendar statistics, split test and RSST variants were rerun the same way. We also corrected the claim that the window starts in November 1992 because HAA and ADM start there: their cards now start in 1974 and 1986. The chart from the May run was removed because it showed the old figures.

Update, 5 October 2026: the financing paragraph now says that since 5 October 2026 our backtests charge swap-based funds their filed spread on the full notional. No figure in this post changed.

Correction, 8 October 2026: three sentences quoted HAA Leveraged 2x's 23% CAGR, or the drawdowns of the aggressive strategies, without saying they come from the 1992 to 2026 window of the results table. They now say so and give HAA Leveraged 2x's full card, which starts in 1974: 26.4% a year and a -36.5% maximum drawdown. The bottom line also put the aggressive strategies' worst drawdowns at roughly 30 to 35%; HAA Leveraged 2x reaches -36.5%, so the range is now roughly 25 to 37%. We rechecked the table, regimes, calendar statistics and split test on each card's latest run, and none of them changed.

Correction, 12 October 2026: the post described the wrong Cockroach. BestFolio's Cockroach is Tony Dong's: 20% each of XLP, XLU, XLV, GOVT and GLD, rebalanced quarterly, with no cash and no managed futures, so the section on a cash leg was removed and the 2022 paragraph rewritten (none of the five sleeves made money that year). TQQQ Trend reads SPY's 200-day average, not QQQ's; ADM compares SPY with SCZ; HAA's canary is TIP's absolute momentum and its 2x variant holds 2x funds; the Quartet runs about 1.55x. The figures are unchanged.

Correction, 13 October 2026: the synthetic history of GDE before its March 2022 launch now matches the fund's 90/90 exposure (90% S&P 500 and 90% gold futures), so the Quartet's card was rerun: on this window its return moved from 11.51% to 11.34% a year and its maximum drawdown from -28.9% to -27.9%. We recomputed the blend from the new Quartet history on the same window and weights. It now returns 13.62% a year (was 13.67%) with a -19.7% maximum drawdown (was -19.8%) and the same 0.90 Sharpe. The regime table, calendar statistics, split test, simplified variant and longest common window were rerun the same way; 2008 is now a flat positive year for the blend, so it has 30 positive calendar years out of 35 rather than 29.

Correction, 15 October 2026: the managed-futures history before KMLM now uses the BTOP 50 index (and DBMF from 2019), so the cards of the Quartet, TQQQ Trend and ADM Stack were rerun, and volatility is now measured on observed prices (months rebuilt from monthly data count at their month-end returns), as on the strategy pages. On this window the Quartet now returns 11.75% a year (was 11.34%), TQQQ Trend 15.36% (was 14.76%) and ADM Stack 16.64% (was 16.24%). We recomputed the blend on the same window and weights: 13.87% a year (was 13.62%), a -20.2% maximum drawdown (was -19.7%) and a 0.94 Sharpe (was 0.90), now above every component rather than level with HAA Leveraged 2x. The summary still showed the 5 October figures (13.67% and -19.8%) and now matches the table. The regime table, calendar statistics, split test, simplified variants and longest common window were rerun the same way.

Correction, 16 October 2026: early-1990s international and emerging-market history now uses index data instead of two actively managed funds, so the cards of HAA Leveraged 2x and ADM Stack were rerun. On this window HAA Leveraged 2x now returns 23.40% a year (was 23.33%) and ADM Stack 16.37% (was 16.64%) with a 0.83 Sharpe (was 0.86). We recomputed the blend on the same window and weights: 13.84% a year (was 13.87%), the same -20.2% maximum drawdown and a 0.93 Sharpe (was 0.94), still above every component. 1994 is now a small positive year for the blend, so it has 31 positive calendar years out of 35, and 1995 rises to +40.0%. The split test, simplified variants and longest common window were rerun the same way.

Correction, 17 October 2026: every card was rerun with longer warm-up histories. On this window ADM Stack now returns 15.77% a year with a -28.3% worst drawdown (was 16.37% and -24.1%). We recomputed the blend on the same window and weights: 13.82% a year (was 13.84%) and a -20.5% maximum drawdown (was -20.2%), now reached in October 2008 rather than March 2020. 2008 is now a small negative year, so the blend has 30 positive calendar years out of 35; 1995 is +38.8% and the worst rolling 12 months -18.9%. On the longest common window, now from January 1985, the blend returned 14.82% a year. The split test and simplified variant were rerun the same way.

Correction, 18 October 2026: the results table still showed ADM Stack's Sharpe and Sortino from before the 17 October restatement (0.83 and 1.21, now 0.78 and 1.14), and three regime rows, the 2022 figure and a few volatility and Sortino cells had not been carried either; they now match the recomputed blend. The Sharpe and Sortino here use this post's own definition, above a 3% rate, so the 18 October switch of the strategy cards to Sharpe measured above cash does not change them, see the changelog.