Trend Following (3/4): What Trend Following Actually Adds to a Risk-Premia Core

Combine a three-asset risk-premia portfolio with a trend-following program and the Sharpe ratio jumps from 1.1 to nearly 1.5. It looks like free diversification. But a trend program is long equities, bonds and metals — and a risk-premia core is equities, bonds and metals. So before we accept the free lunch, we should ask what part of it we are paying for twice, and what part is genuinely new.

The core throughout this piece is RP3 — the inverse-volatility TAA portfolio of US equities, long Treasuries and gold introduced in the earlier IVOL articles, scaled to an eight-percent target. Readers will know it well, and nothing here changes its definition. The held-asset analysis runs on my own synthetic total-return ETF series — SPY, TLT and GLD — which I keep labelled as ETF tickers throughout precisely to hold the held-asset core visually distinct from the futures trend program. The trend program is a 62-market futures system spanning ten sectors; to keep the comparison tradeable I use not the full universe but a per-sector elastic-net replica — roughly ten instruments, one representative cluster per sector, refit on a rolling window. All figures begin in 1980 where the futures history allows; I lean on the 1995-onward window as the more representative read, because the trend universe is thin before then. The selection here is a rolling walk-forward refit, so I make no out-of-sample claim — with the later history already known, none of this is a true holdout; it is a like-for-like construction across a cleaner period.

The hook: two 1.1-Sharpe strategies that combine to 1.5

Here is the naive starting point. RP3 alone earns a Sharpe of about 1.1 over 1995–2026. The trend replica, standalone, earns almost exactly the same. Blended fifty-fifty on equal risk, the pair earns 1.49, and the worst drawdown shrinks from seventeen percent to thirteen. The reason is visible in one number: the two return streams correlate at 0.07. Almost everything good here is the low correlation, not the individual Sharpes.

RP3 and a per-sector EN replica of the trend program each earn ~1.1 Sharpe standalone; blended 50/50 at correlation 0.07 the combination reaches 1.49, with a shallower drawdown. The lift is the diversification, not the components.

This is where most write-ups stop — declare victory and move on. The uncomfortable question is whether a correlation of 0.07 is real diversification or an accounting artifact, given that both books ultimately hold the same kinds of risk.

We already hold these assets

Split the trend program into two halves: the markets the core already holds — equities, bonds and metals, which I will call EBM — and the complement, the markets it does not, namely currencies, energy, grains, softs, rates and the rest. When each half is correlated against RP3, the result is stark. The complement sits at 0.01, essentially orthogonal. EBM sits at 0.14 — modest, but not nothing. The diversification that powered the hook lives overwhelmingly in the markets the core does not own.

The complement (markets RP3 does not hold) is near-orthogonal and supplies almost all the blend Sharpe; the held classes contribute a smaller, real amount. The free lunch is mostly the things you do not already own.

This already reframes the exercise. Trend following is not adding a magic overlay to our assets; it is mostly adding exposure to a different set of markets entirely. The interesting residual is that 0.14 from EBM — small, but worth understanding, because it tells us what trending our own assets is and is not good for.

Why the held part is modest: long is redundant, short is the hedge

Decompose the EBM sleeve into its long and short legs. The long leg — being long equities, bonds or metals when they are trending up — correlates 0.63 with RP3. That is mechanical: a trend system long a rising asset is holding the same exposure the core already holds. As a diversifier it is dead weight. The short leg is the opposite: it carries the crisis behaviour, the payoff that arrives when those same assets fall hard.

The EBM long leg (correlation +0.63 to RP3) is mechanically redundant with simply holding the assets; the short leg carries the crisis payoff, spiking in 2008 and 2022. The diversification is in the short side.

It would be tempting to discard the long leg as pure redundancy. That would be a mistake, and the reason is skew. Trending the held assets — even on the long side — lifts the monthly skewness of the combined book from roughly zero to strongly positive. The long leg earns its place not through correlation but through the shape of the distribution: it trims the left tail and stretches the right, which for a compounding investor is worth more than the Sharpe arithmetic suggests.

Adding the per-asset trend to RP3 lifts monthly skew from +0.06 to +0.91. The long leg is redundant on correlation but reshapes the distribution — the reason to keep it.

What does the core actually need protected?

If the valuable part of the held-asset trend is the short side, the natural question is what RP3 needs protecting from. Here is its left tail at two frequencies. The daily conditional shortfall — the average of the worst five percent of days — is about one percent; the monthly figure is near four. Those bad months are not single-day crashes but grinding, multi-week declines where equities, bonds and gold fall together. That is the shape a hedge has to answer.

RP3 left tail: daily and monthly CVaR-5%. The damaging episodes are multi-week declines, not single-day shocks — which is what the short side of trend is built to catch.

Where the tail cover actually comes from

Before reaching for the held-asset short, it is worth asking which sectors already cover those bad days. Conditioning every sector’s trend sleeve on RP3’s worst five percent of days gives a clear picture. The complement sectors — energy, grains, crypto, short rates — lean positive when the core is bleeding, even though none of them is a directional hedge; their orthogonality simply means they are doing their own thing while equities, bonds and gold fall together. The held-asset trend, by contrast, actually loses on the worst single days, because trend is slow and is still long the asset that is falling.

This sharpens the earlier claim. The complement is not merely a return diversifier that happens to sit out the crisis — on a typical bad day it is the part of the book leaning the right way. The held-asset short is a different animal: it pays in sustained crises, the 2008s and 2022s, not on the worst individual days. Two distinct hedges with two distinct timescales.

Per-sector trend contribution on RP3’s worst-5% days. The complement (green) leans positive — energy, grains, crypto, rates cover the day-to-day tail; the held-asset trend (navy) loses on the worst days, being slow to flip. Sustained-crisis cover is a separate job, done by the short legs.

Realizing the hedge on our own tickers

So we apply the full trend signal directly to RP3’s own three tickers — equities, long bonds and gold — rather than to a separate futures program. The long side reproduces the futures EBM almost exactly, as it must. The short side tracks it too, but with two conspicuous gaps.

Per-asset trend on SPY/TLT/GLD reproduces the EBM long leg in full; the short leg tracks the futures EBM but misses the 2008 and 2022 magnitudes.

Decomposing those gaps by sector shows exactly where the ticker-level book falls short. In 2008 the crisis hedge came from shorting industrial metals as the economy seized up — copper above all. RP3’s metal is gold, a safe haven that rose into the panic, so there was simply no short to take. In 2022 the hedge came from shorting the whole rate curve; RP3 holds long-duration Treasuries, so it caught the long end but underweighted the front. The two missing exposures are an industrial metal — copper is the textbook example — and the short end of the curve.

The short-side gap by sector. In 2008 the EBM hedge is almost entirely industrial metals (copper); RP3’s gold cannot reproduce it. In 2022 the bond hedge matches; the residual is the front of the curve.

Closing the gap: the minimum add-on

The fix is to build the missing piece the same way every sector is built. We take the short side of the EBM trend over the cleaned universe — industrial metals after dropping gold and silver — silver too, because it is far too correlated with gold to act as an industrial crisis short rather than a second safe haven — the curve after dropping the long bonds the core already owns — and run the identical elastic-net replica we used for the seven complement sectors: select which instruments reproduce the sector’s own short sleeve, equal-weight the chosen, refit on a rolling window, lag the selection. The only differences from a normal sector are that the legs are short-only and that the position is sized off the volatility of its full trend leg rather than the intermittent short stream, so the rare shorts are never silently levered. The replica rotates through copper, platinum and palladium on the metals side and the five- and ten-year on the curve, exactly as the data warrants each quarter. Standalone it bleeds gently — the premium you pay for protection — but it is negatively correlated with the core (−0.29), so fifteen percent of it on RP3 lifts the Sharpe and trims the drawdown rather than the reverse. It is insurance you are paid to hold.

With this add-on, the reconstruction of the futures crisis hedge from RP3-accessible pieces falls into place. On its own, the per-asset short of the held tickers correlates only 0.58 with the full program’s EBM short and is essentially blind to 2008 — the clearest proof that trending your own assets does not rebuild the trend short. Adding the EBM-short replica lifts the correlation to 0.76 and recovers the 2008 episode the ticker book missed entirely. That is the minimum addition that closes the gap.

RP3’s own ticker shorts reconstruct only part of the EBM trend short (0.58, blind to 2008). The EN replica of the industrial-metals and front-curve short — built like every other sector — lifts the reconstruction to 0.76 and recovers 2008. The minimum add-on that closes the gap.

The build

Assembled, the strategy has two legs, each at equal risk. The first is the enhanced core: RP3 plus the per-asset trend on its own tickers, which supplies the long-side skew and de-risks the held assets when they fall. The second is the orthogonal trend: the complement sectors that RP3 does not own, plus the EBM-short add-on — an elastic-net replica of the industrial-metals and front-curve short legs, built with the same recipe as the complement sectors. Nothing in the trend sleeve is redundant with the core — no held-asset long exposure beyond what the per-asset tilt already provides, and no equities sector at all. The two legs correlate at 0.04.

The two-leg build: RP3 plus per-asset ticker trend (Leg A) and the orthogonal complement plus crisis shorts (Leg B), 50/50 vol-balanced to 8%. The portfolio rides above both legs with a consistently shallower drawdown than RP3.

Performance across windows, gross of costs. The 1995–2026 row is the representative read; the strong 1980s trend era lifts the full-sample figures. RP3 alone shown for reference.

The representative figure is the 1.40 over 1995–2026, not the 1.67 across the full sample, which the unusually strong 1980s trend era lifts. The Sharpe lift over the bare core is regime-dependent and has thinned as trend following’s easy era receded. What does not depend on that regime is the drawdown control, pinned near eleven percent across every window, and the positive skew — both products of the structure rather than of trend being hot.

The combined strategy, with the major crises flagged. It held its worst drawdown near eleven percent across 46 years — about a third less than the core alone — and came through the dot-com bust and the 2008 crisis strongly, the rest roughly flat. Not a perfect hedge in every episode, but resilient where it counts.

One caveat in the same honest spirit: the EBM-short add-on is close to Sharpe-neutral at the portfolio level. The complement already earns much of the 2008 return on its own — short energy, long short-rates, long dollar — so the explicit crisis replica improves the drawdown and completes the EBM-short reconstruction more than it lifts the ratio. It is in the build because it closes the gap and hardens the tail, not because it manufactures Sharpe.

A word on what these numbers are and are not. Every figure in this piece is gross of costs: no commissions, no slippage, no bid–ask, no management fee. The only frictional input is financing on de-risked cash, accrued at the Fed Funds rate. Turning this research portfolio into a net, implementable mandate — the turnover it generates, the instruments you would actually trade, the financing and custody realities — is the subject of the closing article in the series.

The limits

The honest close is that the orthogonality is not absolute. In late 2016 both legs fell together: a sharp post-election reversal in rates whipsawed the trend sleeve while the same move dragged the core’s bond exposure. A shared rates shock is the one factor both books can be exposed to at once, and it produced the deepest drawdown in the sample.

The combination is not perfect, and there is no point pretending otherwise. But the value of decomposing the program this way is exactly that each gap, once understood, points to the diversifier that would close it — which is how we arrived at the complement and the EBM-short add-on in the first place. Filling the remaining gaps is a matter of understanding them, not of bolting on more complexity; the structure tells you what is missing.

The one episode the orthogonality does not cover: the 2016 post-election rates whipsaw dragged both legs down together, the deepest drawdown in the sample — a reminder that the combination is not perfect, and a pointer to where the next diversifier belongs.

What trend adds, precisely

Trend following adds value to a risk-premia core in two separable ways: the markets you do not hold, which supply genuine diversification and, on the core’s worst days, already lean to the hedging side; and the short side of the markets you do, which supplies the deep-crisis payoff that arrives in the sustained declines. The long side of the held markets is not diversification — it is the assets you already own, kept only for the positive skew it lends the distribution. Crucially, the core’s own tickers do not rebuild that short on their own — they are blind to the 2008 industrial-metals crash and underweight the front of the curve — so the minimum addition is a short-only EN replica of exactly those missing EBM legs, built like any other sector, which the negative correlation makes nearly free to carry.

The result is a portfolio that holds its worst drawdown near eleven percent across four decades and keeps a positive return skew throughout — not by predicting crises, but by owning the right orthogonal exposures and refusing to pay twice for the ones already in the book.