SI410: The Next Evolution of Trend Following ft. Nick Baltas

State of the trend book. Baltas frames 2026 as a good year for trend following: five of seven months positive, excluding a March V-shape reversal and a June drawdown driven mainly by equities. July's rebound (as of Wednesday) has been "primarily driven by rates activity in the last week," with commodities also contributing both month- and year-to-date; equities and FX are roughly flat to slightly down MTD. Carry has been a standout diversifier this year in both FX and commodities, the latter aided by backwardation tied to geopolitical activity. Levels cited: BTOP50 +1.34% MTD/+10% YTD; SocGen CTA Index +0.89%/+10.37%; SocGen Trend Index +94bp/+10.15%; Short-Term Traders Index -0.6%/+4.52%; MSCI -84bp/+9%; US Aggregate Bond Index -96bp/-12bp YTD. Nothing this year has surprised him outright, but he notes March's drawdown was unusually shaped — "it was more of a fast recovery rather than like a fast drop" — making it less impactful for trend than prior V-shapes, and that dynamic volatility scaling work his team has done "played out quite nicely" in that environment.

How many markets should a trend program trade? A new paper from Graham Capital ("An Expanding Market Universe," Nashja Betker and Thomas Feng) tests progressively adding less-liquid markets to a core universe and finds the incremental benefit largely disappears past roughly 70–80 markets — but only for macro instruments (equities, currencies, rates), where added names are already spanned by existing principal components. Less-liquid, more exotic commodities (the paper splits Chinese vs. non-Chinese commodities) do add incremental long-term performance even past 100 markets, though not linearly, since liquidity costs create a "hump." Baltas says this validates his own range: his programs run 50–100 markets, settling near 20 per asset-class bucket, and adding more within an asset class simply dilutes each market's risk contribution without adding diversification.

Defensive architecture and portable alpha. A separate paper, "The Perfect Hedge" (One River Capital, Patrick Casley), sorts portfolio protection into three buckets: first responders (options/volatility hedges — low basis, high cost, reactive to flash corrections), second responders (trend following, capturing multi-month crisis alpha rather than short shocks), and a third, non-responder bucket of diversifying, alternative-risk-premia exposure with flat-to-negative equity correlation that moderates the carry cost of the other two. Baltas agrees with the framework and stresses one design constraint: the third-bucket diversifier "cannot be a short gamma exposure," or the portfolio simply recreates the downside risk it's trying to hedge in order to fund it — at that point it's a relative-value trade, not a defensive overlay.

On portable alpha — stacking trend on top of an existing book — Baltas's answer to concerns about adding leverage depends on what's underneath. For an equity-only base, he'd size the trend overlay to reduce upside equity participation so the combined book doesn't double up on equity risk on the way up. For an equity/bond base, he argues for a more unconstrained trend profile, because the real tail risk in that portfolio is an inflation shock that drives equities and bonds down together — a regime that "happens over the course of a quarter if not more," which is exactly the multi-month macro shock medium-term trend is built to capture. His framing: the leverage itself isn't the risk; concentrated leverage on the asset you're already overexposed to is.

Thematic and narrative investing. Baltas discusses two research strands from a recent industry conference. The first, "Thematic Investing as Missing Factors" (Wiley Lee), argues that conventional risk models — built on fundamental characteristics and categorical tags like sector or country — miss thematic clustering (e.g., AI-exposed names), and that ignoring it leaves realized portfolio risk higher than the model predicts. The second, a "narrative momentum" paper, builds a dataset of roughly 13.3 million articles, corporate filings, trading blogs and news items since around 2013–14, structured into 347 narratives (with a robustness check using 57 narratives drawn from academic JL classifications), and finds that stocks with rising exposure to strengthening narratives outperform those tied to fading ones — a signal distinct from, but related to, price momentum, pointing to an underlying demand mechanism behind trends themselves.

Baltas's own framing is that thematics are fundamentally a dimensionality-reduction exercise, akin to sector or country tagging, not necessarily a standalone risk premium — some categorical exposures carry compensation, most don't, but all affect realized correlation as capital allocates by narrative rather than by asset. On the backtesting problem — you can't know a theme (AI, the Strait of Hormuz) before it exists — Baltas's answer is to rely on a smaller set of "evergreen" narratives (roughly 30–60, e.g., trade wars, stagflation, capex, geopolitical tension) that predate any specific event and can absorb new instances of the same underlying story, an approach he compares directly to managing the equity factor zoo. On whether this is simply repackaged sector investing dressed up as something new, Baltas partly concedes the point: the underlying truth is still price and macro regime, and thematic tagging is primarily a way to compress information overload for human (particularly discretionary) decision-making rather than a newly discovered source of return.