Replicating CTA Trend Following & Crisis Alpha

The Retirement Case: What Broke in 2022

The Correlation You Are Implicitly Short

A 60/40 portfolio is not two assets. It is one asset plus a bet that the stock-bond correlation stays negative. For roughly the two decades before 2022, that bet paid: equity drawdowns were disinflationary shocks, the Fed cut, bonds rallied, and the bond sleeve did exactly the job it was hired to do. A retiree drawing down through 2000-2002 or 2008 could sell bonds at a gain to fund withdrawals rather than liquidating equities at the bottom.

In 2022 that relationship inverted. The shock was inflationary, not deflationary, so the discount rate rose against both legs simultaneously. Global equities and long-duration Treasuries fell together, and the diversification a retiree had been paying for in expected return — decades of accepting bond yields below equity earnings yields — simply did not arrive at the moment it was needed.

The Framing That Governs Everything Below

Trend following is a hedge sleeve and a diversifier. It is not a return engine, and it is categorically not a substitute for owning the market. Over long horizons, a low-cost total-market index core is what actually funds a retirement; every strategy in this article is layered on top of that core to change the path the core travels, not to replace it.

The practical form of that statement: a 5-15% allocation, funded from the bond sleeve rather than from equities, rebalanced mechanically, and evaluated on whether it reduced portfolio drawdown — never on whether it beat the S&P 500 in any given year. If you find yourself comparing this sleeve's annual return to your index fund's, you have already misunderstood what it is for, and you will sell it at the worst possible time.

Time-Series Momentum: The AQR Century Study

Cross-Sectional vs. Time-Series Momentum

Time-series momentum asks a different question from the factor momentum covered elsewhere in this series. Cross-sectional momentum ranks assets against each other and buys the winners relative to the losers — it is market-neutral by construction and says nothing about the direction of the market as a whole. Time-series momentum (TSMOM) asks only whether each individual market's own past return was positive or negative, and takes a long or short position in that market accordingly, independent of what every other market is doing.

That distinction is the whole reason TSMOM produces crisis alpha and cross-sectional momentum does not. A cross-sectional book stays roughly dollar-neutral through a crash. A time-series book, when every market it trades has been falling for months, ends up net short across the entire portfolio — which is exactly the position you want during a prolonged bear market and exactly the position no other component of a retirement portfolio will take for you.

What "A Century of Evidence" Actually Established

The AQR research program associated with Tobias Moskowitz, Yao Hua Ooi and Lasse Pedersen — the "Time Series Momentum" line of work and its later extension into a multi-century backtest — made one claim that matters more than all the performance statistics: the effect is not a product of the sample. The same simple signal, applied to whatever liquid futures and forwards existed in each era, produced positive risk-adjusted returns across the interwar period, the post-war era, the 1970s inflation, the 1980s disinflation, and the modern period.

That is a materially stronger result than the usual factor backtest. Most documented anomalies are established on U.S. equity data since 1963 and quietly decay out-of-sample. TSMOM's evidence base spans multiple asset classes, multiple monetary regimes, and a period long enough to include events that no living portfolio manager traded through. It is the closest thing in systematic investing to a genuinely out-of-sample confirmation.

The Behavioral and Structural Explanations

Trend persistence is usually attributed to some combination of slow information diffusion, investor under-reaction followed by delayed over-reaction, and — importantly for a retirement investor — the existence of non-profit-seeking participants who are structurally obliged to trade in the direction of the move. Central banks smoothing currency adjustments, corporates hedging commodity exposure on a schedule, pension funds rebalancing to policy weights, and risk-control mandates forcing de-leveraging into a decline all supply flow that a trend follower is on the other side of.

This matters because it suggests the source of return is not purely a behavioral bias that arbitrage should compete away. Some of it is a genuine risk-transfer premium: the trend follower is providing liquidity to participants who must trade regardless of price. That component has more reason to persist than a pure anomaly does — though "more reason to persist" is a long way from "guaranteed."

Man AHL: Alpha, Alternative, or Arithmetic?

The Question the Industry Would Rather Not Ask

Man AHL's research framing — is trend following alpha, an alternative asset class, or arithmetic? — is the most useful piece of intellectual honesty in the managed futures literature, because the answer determines what a fair fee is. If trend following is alpha, it is manager skill, it is scarce, and 2-and-20 might be defensible. If it is an alternative asset class, it is a risk premium anyone can harvest and it should cost what a factor ETF costs. If it is arithmetic — a mechanical consequence of a rule applied to price series — then the fee should approach the cost of running the arithmetic.

The uncomfortable answer that emerges from the research is: mostly the third, with a modest layer of the first. The bulk of a typical CTA's return can be replicated by a simple, published, mechanically-specified trend rule applied to liquid markets. The residual — market selection, execution quality, risk overlays, faster or adaptive signal blends — is real but small relative to the headline fee.

The Fee Arithmetic Is Brutal

If the underlying strategy delivers a long-run Sharpe ratio in the 0.4-0.7 range at a 10% volatility target — a fair reading of the published evidence net of realistic costs — then a 2% management fee plus 20% performance fee consumes a large fraction of the risk-adjusted return. At 10% vol and a 0.5 gross Sharpe, gross excess return is roughly 5%; a 2% management fee alone removes 40% of it before the incentive fee touches anything.

What This Means for the Build-vs-Buy Decision

The honest framing is a three-way choice, and each option has a real cost. A DIY 5-ETF replication has near-zero management fee but the narrowest breadth and the highest behavioral risk (you can turn it off). A liquid managed-futures ETF has moderate fees and better breadth but often significant strategy drift between products. A managed futures mutual fund has the best breadth and the worst fees. There is no dominant answer; the point is to make the trade-off explicitly rather than defaulting to whichever product happens to be marketed to you.

Volatility-Targeted Sizing: Where the Sharpe Actually Comes From

The Sizing Rule

In a TSMOM program, position size in each market is set inversely proportional to that market's recent realized volatility, so that every market contributes approximately the same risk to the portfolio regardless of how volatile it happens to be. A position in a 6%-volatility bond future and a position in a 45%-volatility natural gas future are sized so that a one-standard-deviation move in either produces a comparable P&L impact.

Position size for market i at time t:

    w_i,t  =  sign(signal_i,t)  ×  (σ_target / σ_i,t)  ×  (1 / N)

where:
    sign(signal_i,t)  = +1 if trailing return positive, -1 if negative
    σ_target          = per-market annualized volatility budget (e.g. 10%)
    σ_i,t             = trailing realized volatility of market i (e.g. 30-day,
                        annualized, often exponentially weighted)
    N                 = number of markets in the universe

Portfolio-level scaling (applied after the per-market weights):

    scale_t = σ_portfolio_target / σ_realized(portfolio_t)

    final weights = w_t × scale_t     (capped at a leverage limit)

Realized volatility, 30-day annualized:

    σ_i,t = sqrt(252) × stdev( r_i,t-29 ... r_i,t )

Exponentially-weighted alternative (faster to react, less noisy on the tail):

    σ²_i,t = λ · σ²_i,t-1 + (1 - λ) · r²_i,t        with λ ≈ 0.94

The inverse-volatility term is doing more work than the signal. This is the counterintuitive result that most retail implementations miss: research on trend following consistently finds that a substantial share of the risk-adjusted improvement over a naive equal-weight trend portfolio comes from the volatility scaling rather than from any refinement of the entry signal. Investors spend enormous effort optimizing lookback windows and almost none on sizing, which is exactly backwards.

The Feedback Loop That Creates Crisis Convexity

Volatility targeting interacts with trend signals to produce a return profile that resembles a long-straddle payoff — and that resemblance is not accidental. When a market is calm and trending, realized volatility is low, so the inverse-vol rule takes a large position in the direction of the trend. When volatility explodes, the rule cuts position size automatically, which caps the damage from a sudden reversal. The combination — big positions in slow, persistent moves; small positions in chaotic ones — is precisely the exposure profile that pays during extended crises and bleeds during choppy ones.

Crisis Alpha: 1970s, 2008, 2022

The Mechanism, Stated Precisely

Crisis alpha is not a claim that trend following goes up when stocks go down. It is a claim about a specific type of crisis: one that unfolds over months rather than days, and that moves multiple asset classes in a persistent direction. Trend following requires time to establish positions. A signal based on trailing returns is, by construction, late — it cannot be short before a decline starts, only after the decline has persisted long enough to flip the signal. The strategy therefore monetizes the second half of a long drawdown, not the first shock.

Case 1: The 1970s Stagflation

The 1970s are the single most important case study because they are the regime a modern portfolio is least prepared for and the one 2022 rhymed with. Through 1973-1974 and again in 1979-1980, inflation and rate expectations rose persistently, commodities rose persistently, the dollar moved persistently after the Bretton Woods breakdown, and equities and bonds both delivered deeply negative real returns. Every one of those was a multi-month directional move in a liquid futures market — the textbook environment for time-series momentum.

A trend program in that decade would have been long energy and agricultural commodities, short bond futures as yields rose, and positioned in currencies against the dollar's trend — all simultaneously, and all in the same direction as the forces destroying a conventional portfolio's real value. This is the historical episode that the AQR century-length backtest was specifically designed to include, and it is why the multi-century framing matters: an evidence base that starts in 1990 has never seen a genuine inflationary regime.

Case 2: 2008

2008 is the cleanest modern demonstration and also the source of most of the category's subsequent overselling. The crisis was slow enough to be tradeable: credit stress built through 2007, commodities ran to a mid-2008 peak and then collapsed, equities declined over more than a year, and bond futures rallied hard as policy rates went to zero. A diversified trend program had time to rotate — long commodities into the peak, then short them into the collapse; short equities through the decline; long bonds through the rally.

Managed futures as a category posted strongly positive returns in 2008 while global equities fell sharply, and that single year did more for CTA asset-gathering than the preceding twenty. It also set an expectation the category could not meet: investors who bought after 2008 were buying a hedge at its most expensive moment, in terms of both valuation of the narrative and the subsequent decade of performance.

Case 3: 2022

2022 is the most relevant case for a current retirement portfolio because it is the year the standard diversifier failed and trend following did not. The setup was almost purpose-built for TSMOM: a persistent, telegraphed rate-hiking cycle producing one of the worst years in the history of long-duration Treasuries; a persistent dollar uptrend against the yen, euro and sterling; a persistent commodity move driven by energy dislocation; and a grinding equity downtrend without a single decisive crash.

Four asset classes, all trending, all for most of a year. Managed futures indices posted strongly positive returns while a 60/40 portfolio suffered one of its worst calendar years on record. For a retiree drawing down in 2022, a trend sleeve was one of the only line items in the portfolio that could be sold at a gain to fund withdrawals — which is the entire practical value of crisis alpha stated in one sentence.

The Drought: 2011-2019 and Why It Nearly Killed the Category

What a Decade of Whipsaw Feels Like

Between roughly 2011 and 2019, diversified trend following delivered close to nothing — and in several individual years, meaningfully less than nothing — while a 60/40 portfolio compounded through one of the strongest bull markets on record. The environment was structurally hostile: central bank intervention repeatedly truncated developing trends, equity drawdowns were sharp and immediately reversed rather than persistent, rates were pinned near zero so bond trends had little room to run, and cash yielded nothing, so the collateral return that historically padded CTA performance disappeared entirely.

The result was a strategy that repeatedly established positions in the direction of a nascent move, then had that move reversed by a policy announcement, then flipped and got reversed again. Each cycle costs the transaction spread plus a small loss. Do that dozens of times a year across a portfolio and the drag is severe.

The Behavioral Problem Is Worse Than the Financial One

Nine years of flat-to-negative performance while your index fund triples is not a statistical inconvenience — it is a psychological ordeal that most investors do not survive. The pattern is documented and depressingly consistent: investors capitulate on the diversifier late in the drought, and the money that leaves at the bottom is not there for the crisis that follows. Investors who exited managed futures in 2020-2021 after a decade of disappointment missed exactly the year the sleeve existed for.

The 5-ETF Retail Replication

The Universe and Why These Five

The retail replication compresses a 60+ market futures universe into five liquid ETFs, one per major trend-following asset class, chosen for liquidity and for representing a distinct macro driver. This is a crude approximation and the section after this one is entirely about how crude — but it is the version an ordinary brokerage account can actually run, in a tax-aware way, without margin, without futures approval, and without a Bloomberg terminal.

The Long/Flat Constraint

The single biggest structural difference between the retail replication and the institutional strategy is that a retail sleeve is typically long-or-flat, not long-or-short. A CTA that flips short bond futures in 2022 earns the full decline. A retail replication that simply exits TLT and sits in T-bills avoids the decline but does not profit from it. That halves the crisis alpha — and it is a deliberate, defensible trade-off, because shorting via inverse ETFs introduces daily-reset decay, and shorting via futures requires an account type and a risk tolerance most retirement investors should not have.

Cash Yield Optimization: The Part Everyone Skips

In a long/flat replication, a large fraction of the time some sleeves are flat — and where that cash sits determines a surprisingly large share of the strategy's total return. This is the retail echo of the institutional collateral return: a CTA posts margin and earns T-bill rates on the rest, and in a 4-5% rate environment that collateral yield can be a substantial part of the headline number. A retail replication that leaves flat sleeves in a zero-yield sweep account is discarding the same contribution.

The Rebalance Rule

Monthly, on the last trading day: for each of the five sleeves, compare the price to its 10-month simple moving average (equivalently, roughly the 200-day average). Hold the ETF if price is above; hold cash if below. Then apply inverse-volatility weights across the held sleeves and scale the whole sleeve to its target volatility. Monthly evaluation is not a compromise — the research on trend following consistently finds that monthly and daily specifications produce broadly similar results, and monthly dramatically reduces both transaction costs and the number of opportunities you have to override the system.

Python Implementation: Vol-Targeted TSMOM Engine

Full implementation: Realized volatility estimation, TSMOM signal generation with a hysteresis buffer, inverse-volatility position sizing, portfolio-level volatility targeting, cash-yield accounting on flat sleeves, and a monthly-rebalanced backtest with turnover-based transaction costs.

Complete Python Code

"""
Volatility-Targeted Time-Series Momentum (TSMOM) Engine
Retail 5-ETF replication of the CTA trend-following return profile,
following the AQR (Moskowitz, Ooi & Pedersen) TSMOM framework and the
Man AHL 'alpha vs. arithmetic' replication critique.

Long/flat only (no shorting) - a deliberate retail constraint that
bounds downside at the cost of roughly half the crisis alpha.

Author: Plan My Retire
Date: September 2026

Requires: pandas, numpy. Price data: any daily adjusted-close source.
"""

import numpy as np
import pandas as pd


# ---------------------------------------------------------------------
# 1. Volatility estimation
# ---------------------------------------------------------------------

class VolatilityEstimator:
    """
    Trailing realized volatility, annualized. Two estimators:
    a simple rolling window (transparent, what the papers use) and an
    exponentially-weighted variant (reacts faster to regime shifts).
    """

    TRADING_DAYS = 252

    def __init__(self, window=30, lam=0.94):
        self.window = window          # 30-day realized vol per the spec
        self.lam = lam                # EWMA decay; 0.94 is the RiskMetrics default

    def rolling(self, returns):
        """Simple rolling-window annualized volatility."""
        return returns.rolling(self.window).std() * np.sqrt(self.TRADING_DAYS)

    def ewma(self, returns):
        """
        Exponentially-weighted annualized volatility.
        sigma^2_t = lam * sigma^2_{t-1} + (1 - lam) * r^2_t
        """
        var = returns.pow(2).ewm(alpha=1 - self.lam, adjust=False).mean()
        return np.sqrt(var * self.TRADING_DAYS)


# ---------------------------------------------------------------------
# 2. Trend signal
# ---------------------------------------------------------------------

class TrendSignal:
    """
    Long/flat TSMOM signal from a moving-average filter with an optional
    hysteresis buffer to suppress whipsaw round-trips.

    buffer_pct=0.02 means: enter when price is 2% above the MA, exit when
    price is 2% below it. Between those bounds the prior state persists.
    """

    def __init__(self, ma_window=200, buffer_pct=0.02):
        self.ma_window = ma_window    # 200 trading days ~ 10 months
        self.buffer_pct = buffer_pct

    def generate(self, prices):
        """
        prices: DataFrame of daily adjusted closes, one column per ETF.
        Returns a DataFrame of {0, 1} positions, forward-shifted by one
        day so a signal computed on day t is only tradeable on day t+1.
        """
        ma = prices.rolling(self.ma_window).mean()
        upper = ma * (1 + self.buffer_pct)
        lower = ma * (1 - self.buffer_pct)

        raw = pd.DataFrame(np.nan, index=prices.index, columns=prices.columns)
        raw[prices > upper] = 1.0     # decisively above -> long
        raw[prices < lower] = 0.0     # decisively below -> flat

        # Inside the band, carry the previous state forward (hysteresis).
        state = raw.ffill().fillna(0.0)

        # Shift to avoid look-ahead bias: today's close cannot be traded today.
        return state.shift(1).fillna(0.0)


# ---------------------------------------------------------------------
# 3. Position sizing
# ---------------------------------------------------------------------

class InverseVolSizer:
    """
    Sizes each sleeve inversely to its own realized volatility, then
    scales the whole portfolio to a target volatility.

    This is where most of the risk-adjusted improvement over a naive
    equal-weight trend portfolio comes from - not from the signal.
    """

    def __init__(self, sleeve_vol_target=0.10, portfolio_vol_target=0.10,
                 max_sleeve_weight=0.40, max_gross_leverage=1.0):
        self.sleeve_vol_target = sleeve_vol_target
        self.portfolio_vol_target = portfolio_vol_target
        self.max_sleeve_weight = max_sleeve_weight
        self.max_gross_leverage = max_gross_leverage   # 1.0 = unlevered

    def size(self, signals, vols):
        """
        signals: DataFrame of {0, 1} long/flat states
        vols:    DataFrame of annualized realized vols, same shape
        Returns target portfolio weights per sleeve.
        """
        # Inverse-vol scaling, guarding against divide-by-zero on stale data.
        inv_vol = self.sleeve_vol_target / vols.replace(0, np.nan)
        raw_weights = signals * inv_vol

        # Cap any single sleeve so one low-vol asset cannot dominate.
        raw_weights = raw_weights.clip(upper=self.max_sleeve_weight)

        # Normalize gross exposure to the leverage cap. Sleeves that are
        # flat contribute zero, so gross exposure falls in risk-off periods
        # and the residual is held in cash (earning the bill rate).
        gross = raw_weights.sum(axis=1)
        scale = np.minimum(1.0, self.max_gross_leverage / gross.replace(0, np.nan))
        weights = raw_weights.mul(scale.fillna(0.0), axis=0)

        return weights.fillna(0.0)

    def apply_portfolio_vol_target(self, weights, returns, lookback=60):
        """
        Second-stage scaling: measure the realized volatility of the
        strategy itself and scale toward the portfolio target. Keeps the
        sleeve's risk contribution roughly stable through regimes.
        """
        strat_returns = (weights.shift(1) * returns).sum(axis=1)
        realized = strat_returns.rolling(lookback).std() * np.sqrt(252)
        scale = (self.portfolio_vol_target / realized.replace(0, np.nan))
        scale = scale.clip(upper=self.max_gross_leverage).ffill().fillna(1.0)
        return weights.mul(scale, axis=0)


# ---------------------------------------------------------------------
# 4. Backtest with cash yield and transaction costs
# ---------------------------------------------------------------------

class TrendReplicationBacktest:
    """
    Monthly-rebalanced backtest of the 5-ETF long/flat trend sleeve.

    Explicitly accounts for two things most retail backtests omit:
      1. Cash yield on flat sleeves (the retail analogue of a CTA's
         collateral return - material in a 4-5% rate environment).
      2. Turnover-based transaction costs, which are what actually
         make the whipsaw drought expensive.
    """

    def __init__(self, cost_bps=5.0, rebalance='ME'):
        self.cost_bps = cost_bps      # round-trip cost per unit turnover
        self.rebalance = rebalance    # 'ME' = month end

    def run(self, prices, weights, cash_rate_annual):
        """
        prices:            DataFrame of daily adjusted closes
        weights:           DataFrame of daily target weights
        cash_rate_annual:  Series of annualized short rates (e.g. 3M bill)
        """
        returns = prices.pct_change().fillna(0.0)

        # Rebalance only on month ends; hold weights constant in between.
        rebal_dates = returns.resample(self.rebalance).last().index
        held = weights.reindex(returns.index)
        held = held.where(held.index.isin(rebal_dates)).ffill().fillna(0.0)

        # Asset P&L uses yesterday's weights against today's returns.
        asset_pnl = (held.shift(1).fillna(0.0) * returns).sum(axis=1)

        # Cash yield on the un-invested fraction (the flat sleeves).
        invested = held.shift(1).fillna(0.0).sum(axis=1).clip(0.0, 1.0)
        daily_cash = cash_rate_annual.reindex(returns.index).ffill() / 252.0
        cash_pnl = (1.0 - invested) * daily_cash

        # Transaction cost proportional to turnover on rebalance days.
        turnover = held.diff().abs().sum(axis=1).fillna(0.0)
        costs = turnover * (self.cost_bps / 10000.0)

        net = asset_pnl + cash_pnl - costs
        equity = (1.0 + net).cumprod()

        return pd.DataFrame({
            'asset_pnl': asset_pnl,
            'cash_pnl': cash_pnl,
            'costs': costs,
            'net_return': net,
            'equity': equity,
            'invested_fraction': invested,
        })

    @staticmethod
    def summarize(result, benchmark_returns=None):
        """Headline statistics, including the ones that actually matter here."""
        r = result['net_return']
        years = len(r) / 252.0
        cagr = result['equity'].iloc[-1] ** (1 / years) - 1
        vol = r.std() * np.sqrt(252)
        sharpe = (r.mean() * 252) / vol if vol > 0 else np.nan

        running_max = result['equity'].cummax()
        drawdown = result['equity'] / running_max - 1.0

        stats = {
            'CAGR': cagr,
            'Volatility': vol,
            'Sharpe': sharpe,
            'MaxDrawdown': drawdown.min(),
            'CashContribution': result['cash_pnl'].sum(),
            'CostDrag': result['costs'].sum(),
            'AvgInvested': result['invested_fraction'].mean(),
        }

        # The number that decides whether this sleeve belongs in the
        # portfolio at all: correlation to the core during core drawdowns.
        if benchmark_returns is not None:
            bench = benchmark_returns.reindex(r.index).fillna(0.0)
            stats['CorrelationToCore'] = r.corr(bench)

            stress = bench < bench.quantile(0.05)   # worst 5% of core days
            stats['CorrelationInStress'] = r[stress].corr(bench[stress])
            stats['ReturnInCoreStress'] = r[stress].mean() * 252

        return stats


# ---------------------------------------------------------------------
# Example usage
# ---------------------------------------------------------------------

if __name__ == "__main__":
    # Replace with real adjusted closes from your data source.
    # Universe: equity, duration, gold, energy, dollar.
    tickers = ['SPY', 'TLT', 'GLD', 'USO', 'UUP']

    dates = pd.bdate_range('2006-01-01', '2026-06-30')
    rng = np.random.default_rng(42)

    # Synthetic price paths with differing volatilities, purely so the
    # example runs standalone. Real use: substitute actual price history.
    sigmas = {'SPY': 0.16, 'TLT': 0.13, 'GLD': 0.15, 'USO': 0.35, 'UUP': 0.08}
    prices = pd.DataFrame({
        t: 100 * np.exp(np.cumsum(
            rng.normal(0.0002, sigmas[t] / np.sqrt(252), len(dates))))
        for t in tickers
    }, index=dates)

    returns = prices.pct_change().fillna(0.0)

    # 1. Volatility
    vol_est = VolatilityEstimator(window=30)
    vols = vol_est.rolling(returns).bfill()

    # 2. Signal: 200-day MA with a 2% hysteresis buffer
    signal_gen = TrendSignal(ma_window=200, buffer_pct=0.02)
    signals = signal_gen.generate(prices)

    # 3. Sizing
    sizer = InverseVolSizer(sleeve_vol_target=0.10,
                            portfolio_vol_target=0.10,
                            max_sleeve_weight=0.40,
                            max_gross_leverage=1.0)
    weights = sizer.size(signals, vols)
    weights = sizer.apply_portfolio_vol_target(weights, returns)

    # 4. Backtest, with a 4% cash rate on flat sleeves
    cash_rate = pd.Series(0.04, index=prices.index)
    bt = TrendReplicationBacktest(cost_bps=5.0, rebalance='ME')
    result = bt.run(prices, weights, cash_rate)

    stats = bt.summarize(result, benchmark_returns=returns['SPY'])
    print("Trend sleeve statistics (synthetic data - illustrative only):")
    for k, v in stats.items():
        print(f"  {k:24s} {v: .4f}")

    print("\nAverage invested fraction by year:")
    print(result['invested_fraction'].resample('YE').mean().round(3))

Why Retail Replication Tracks Imperfectly

The Gaps, Ranked by How Much They Cost You

A 5-ETF replication will not track a diversified CTA index closely, and understanding exactly why prevents the two failure modes that follow — abandoning the sleeve when it lags, and over-trusting it when it leads.

The Dispersion Problem Nobody Mentions

Even among professional CTAs, dispersion of returns in any given year is enormous — managers running nominally the same strategy routinely differ by tens of percentage points because of speed, universe and risk-overlay choices. That has an uncomfortable implication for replication: there is no single "trend following return" to track. Your five-ETF sleeve will differ from a CTA index, and the CTA index members differ wildly from each other. Judging your replication against any specific manager's number is meaningless; judge it against whether it did the job — positive returns during your core's drawdowns.

Illustrative Portfolio Impact

The figures below are hypothetical illustrations of how a trend sleeve changes a retirement portfolio's risk profile — they are not forecasts, are not derived from any audited track record, and will vary substantially with the period examined, the ETFs chosen, fees, and taxes. They exist to make the trade-off concrete.

Common Mistakes That Destroy a Trend Sleeve

Your Action Plan

Phase 1: Decide Whether You Need This At All (2-4 Weeks)

Timeline: Before any allocation, establish whether a trend sleeve addresses a risk you actually carry.

1. Confirm the core is right first — a low-cost, broadly diversified index core with an appropriate bond allocation is worth vastly more than any satellite. If the core has high fees or concentration, fix that instead; this article is irrelevant until it is fixed
2. Quantify your sequence risk — how many years of withdrawals sit ahead of you, and what would a repeat of 2022 followed by 2023 do to the plan? If you are 25 and accumulating, the honest answer may be "nothing meaningful," and no satellite is warranted
3. Model the joint-drawdown scenario explicitly using a withdrawal simulator, so the risk you are hedging is a number rather than a feeling
4. Write down, in advance, the drought you are willing to tolerate — a specific number of years and a specific underperformance. If you cannot commit to ten years, stop here

Phase 2: Paper-Run the Replication (2-3 Months)

Timeline: Run the system with no money at risk and observe your own reactions.

1. Build the signal table for SPY, TLT, GLD, USO and UUP — price versus 200-day average with a 2% buffer — and update it monthly on the same day each month
2. Backtest with real data using the Python framework above, across at least 2006-present so the sample spans 2008, the 2011-2019 drought and 2022; test multiple lookbacks and reject the strategy if only one works
3. Record what the system would have told you to do at moments you remember clearly, and note honestly whether you would have followed it
4. Compare against the buy option — price the managed-futures ETFs and mutual funds available to you, and decide whether their breadth is worth their fee versus your replication's zero fee and narrower universe
5. Decide where flat-sleeve cash will live and confirm your broker's actual sweep rate rather than assuming

Phase 3: Fund the Sleeve and Leave It Alone (6+ Months and Onward)

Timeline: Only after Phase 2 gave you a system you understood and did not want to override.

1. Start at the low end — 5% of the portfolio, funded from the bond sleeve rather than from equities, since the sleeve is replacing part of the diversification job bonds were doing
2. Place it in a tax-advantaged account if you have the space, to avoid short-term-gain drag on monthly rebalancing
3. Write the Investment Policy Statement now — target weight, rebalance date, the exact signal rule, and an explicit statement that the sleeve will not be evaluated against the equity core's return
4. Rebalance mechanically on schedule, including the months you disagree with the output — especially those months
5. Review annually against the correct criterion: correlation to the core, and the sleeve's return during the core's worst periods. Not its absolute return
6. Consider scaling to 10% only after a full cycle of demonstrated ability to hold it through underperformance, and never after a year in which it just performed well

Recommended Reading

• Institutional Research:
  • AQR Capital Management — A Century of Evidence on Trend-Following Investing and the underlying "Time Series Momentum" research (Tobias Moskowitz, Yao Hua Ooi, Lasse Pedersen)
  • Man AHL / Man Institute — Trend Following: Alpha, Alternative, or Arithmetic? on strategy replication and fair fees
  • AQR — research on volatility targeting and the option-like payoff profile of managed futures
  • Published work on "crisis alpha" and managed futures behavior during equity drawdowns
• Related PMR Articles:
  • Man Group AHL: Adaptive Trend (companion brief on adaptive lookback windows)
  • AQR Factor Momentum (the cross-sectional counterpart — a different strategy despite the shared name)
  • Capula: Tail Risk Alpha (the convexity-purchase alternative to trend's convexity-mimicking payoff)
  • D.E. Shaw Macro Volatility (regime detection and crisis positioning)
  • Trading Foundations: Risk Management (position sizing fundamentals underpinning everything above)