The Hardest Part of Quantitative Investing

---

Some discussions about quantitative investing focus on the inputs, the data sources. The features or the model architecture. The part that gets less attention is the one that determines whether a quant strategy survives a difficult stretch: the discipline of staying with it when the market is working against you.

A model you can’t articulate is a model you will eventually abandon

A purely statistical model that tells you to go long Europe and short the US for reasons no one can explain works OK until it stops working.

When it stops working, the question becomes whether the model has broken or is simply having a bad quarter. If the underlying logic is a complete black box, the answer could be that you don’t know. That’s when investors pull the plug … usually at exactly the wrong moment.

The version of quant that survives long enough to compound is usually based on  fundamental insights. The fundamentals don’t have to be sophisticated, but they have to be defensible. The single best test of a researcher is the question: what would have to happen for this strategy to fail? Researchers who can’t answer that cleanly have not finished the research.

What has changed in macro investing since the 1990s

The Soros and Druckenmiller era of concentrated macro bets is largely gone. The reason: the data environment changed. Three decades ago, real-time inflation signals came from quarterly government releases. Now they come from web scraping, geospatial data, and language models reading thousands of foreign-language news sources. The information advantage that once produced a few enormous trades a year now produces many smaller signals that are harder to scale into individual positions.

Modern macro investing has become much closer in spirit to high-quality baseball than to high-stakes poker. The goal is to be right slightly more than 50% of the time across many calls, and to compound the small edges into something meaningful over years.

The problem with 60/40 has been overstated for years

People have been declaring the 60/40 portfolio dead for at least a decade. It has had bad stretches, 1994 and 2022 being the most cited. It’s still widely considered the right starting point for many investors and will be for a long time. The bigger change is in what gets layered on top of it.

Institutional portfolios increasingly carry sizable private allocations, and those allocations cannot be traded without accepting a material discount. The conversations multi-asset managers are now having with institutions are less "should we own 60/40" and more "how do we restore strategic alignment given the privates we already hold, while keeping the rebalancing instrument liquid." That second question is the one driving the recent growth in overlays, derivative income strategies, and structured downside protection.

Why long Calls beat long Puts as downside protection

A clean way to hedge a long equity portfolio is not to buy Puts or long volatility outright. Both are expensive in normal regimes, and both can disappoint at the moments investors most need them. A more cost-effective approach is to replace some of the long equity exposure with long-dated Calls. The position behaves like long equity in calm markets and provides convexity when the index drops. The cost shows up as a slightly lower upside, which is the right trade for investors who have already had a long run on the equity side.

Inflation is a risk factor, not yet a recession trigger

The working rule of thumb in modern macro is that equities can do reasonably well with inflation below 4%. Above that level, the math starts to weigh on the multiple, and commodities become much more important to portfolios. The current situation sits closer to the first situation than the second, but the trend matters more than the level.

For investors looking for cleaner inflation exposure, commodities remain the most direct liquid hedge. Real estate and infrastructure provide longer-term hedging. Energy equities benefit during commodity shocks but carry standard equity beta. The most underweighted asset in many multi-asset portfolios is commodities themselves, usually held in smaller weights than the hedging math supports.

Modern policy response is faster than most models were calibrated for

The 2008 financial crisis built slowly over months. Silicon Valley Bank failed in a matter of days. COVID compressed an entire bear market into a few weeks and rebounded just as fast. Models calibrated on slower situations can miss this kind of speed entirely, both on the way down and on the rebound.

The implication is that the most important risk management work in a quant strategy happens at the design stage, not the execution stage. By the time a regime has evolved, the model has usually been wrong. A key question at the design stage is whether the assumptions baked into the parameters would survive a fast policy response. Investors who answer that question early have a better chance of staying with their process when it matters.

How serious quants are using language models

The frontier for quantitative research is language.

Most human knowledge is encoded in text rather than structured data, and the recent generation of language models has expanded what’s extractable. Simpler applications, including sentiment scoring of Fed speeches and earnings calls, are now standard. Newer applications include extracting thematic exposure from analyst reports, product descriptions, and 10-K filings.

One version of this is that language models are statistical tools predicting the next token. They are very good at certain narrow tasks like comparing a current Fed statement to a historical baseline, and they are not magic. Used carefully, with a human reviewing the trades, they offer a real incremental edge. Used without supervision, they offer a faster way to make the same mistakes.

Searching beyond the most crowded trades

Another implication of this data-rich environment is that investors are searching for opportunities outside the more crowded areas of the market. Where that opportunity ultimately emerges matters less than the process itself: combining fundamental judgment, quantitative tools, and the discipline to adapt as conditions change.

The hardest part might be trusting the model

Models are only as useful as the assumptions behind them, and those assumptions have to be questioned. No matter how many decades one may spend building quantitative strategies, successful investing requires judgment, adaptability, and a willingness to revisit your convictions as conditions change. The latter is critical.

One of the more difficult parts of investing is less about building the models themselves and more about maintaining the discipline to follow a process through difficult periods, staying open to new information, and trusting your work when the market is testing your conviction.

---

DISCLAIMER: This article is based on a conversation from Top Traders Unplugged and reflects themes, ideas, and perspectives discussed during the episode. The views expressed are those of the guest and participants in the conversation and should not be interpreted as investment advice or as the official views of Top Traders Unplugged.

To receive future episodes, research, and weekly insights from the world of investing, macro, and trend following, sign up for our newsletter or subscribe on your preferred podcast platform.