For readers interested in the actual tools, without going through the short introduction below, the best place to start is www.geninnov.ai/riskpaper

When Tools Change, So Do Methods

Every generation of investors has worked with the tools of its time. Log tables gave way to calculators, calculators to spreadsheets, and each shift quietly rewrote how portfolios were built. Generative AI is the next such shift. Its effect on research and idea generation is already visible across the industry. We believe its stronger effect will be on risk management.

Risk management has always been the quieter half of portfolio construction. One half of our craft is building conviction. The other half is understanding what those convictions add up to when they sit together in one book, particularly for investors concerned with volatility in their portfolio NAVs on a marked-to-market basis. It is in this second half that the public market investment industry has been working with tools that have not kept pace with the times. 

The original mathematics of portfolio risk, set out by Harry Markowitz in 1952, defined risk as a property of the portfolio itself: how the holdings move together, where diversification genuinely lives. That construct was ahead of its era twice over. The computers of the time could not handle it, and even when the numbers could be produced, no manager could translate them into a story a client could follow, or more importantly, point to the decisions that could be taken to alter the risk characteristics. So the industry simplified. The full mathematics collapsed into single betas, then into standardized factors, then into distance from a benchmark. Each step was reasonable in its time. The cumulative result was not: risk became defined as how different a portfolio looks from an index, and products increasingly resembled the yardsticks they were measured against.

Risk Management for Benchmark Independence

For a fund like ours, this inheritance fails completely. Our portfolio overlaps with almost no index in the world. Run our book against five plausible benchmarks, and you get five different verdicts, none of which describes the portfolio. The numbers measure the yardstick, not the risk.

The purest method for answering the real questions has existed for more than a century. Principal component analysis looks strictly inside the portfolio's own returns. It ignores names, sectors, and stories, and asks the data a single question: which holdings actually move together? The answer is often uncomfortable. A book of 30 stocks can look well diversified on paper and still be, in effect, 3 or 4 large bets wearing 30 costumes. The mathematics strips off the costumes. It finds the baskets a portfolio has built by accident, and shows which positions quietly compound its dominant bets and which genuinely diversify them.

So why has almost nobody used it? Because its answers were unreadable. The method will state, with perfect precision, that a certain hidden factor drives a fifth of your risk. It cannot tell you what that factor is. For decades, even when it became possible for machines to crunch matrices of long time series, its output was a wall of numbers with no story attached, and a risk report the manager cannot narrate is a risk report that does not get used. In other words, computation stopped being a barrier decades ago. Interpretation never did.

That is precisely the kind of barrier generative AI removes. Over the past 2 years, Swaraj Gambhir and team have worked tirelessly on rebuilding our risk process around this idea: recover the true internal structure of the portfolio with pure mathematics, then use AI, under strict constraints, to name each hidden factor in plain language and turn the structure into specific, per-stock decisions. The rule that governs the whole design is simple. The numbers are fully computed; only the names are generated. The AI adds no information and changes no mathematics. It translates. As far as we know, nothing quite like it exists in practice, which is why we are publishing the work in full. My particular thanks to the team for the research paper and white paper that explain, step by step, exactly what we built and why.

Explained at Five Levels of Complexity

The topic is complex, so we have built the explanation at five levels of depth. Choose yours:

• A short video introducing the idea in a few minutes
• An interactive simulation, on a demonstration portfolio, showing how the tool turns numbers into precise actions
• A longer video series for those who want the full walkthrough
• A detailed white paper covering the history of risk measurement and every step we took
• A formal research paper for the technically inclined

All of it lives at paper.geninnov.org, a site we designed specifically for this work, so that the simulation runs properly in your browser and the papers and videos sit in one place. If you would rather start from a familiar address, the same material is introduced at www.geninnov.ai/riskpaper, though the simulation and full detail will still take you to the dedicated site. 

Two honest caveats belong in even a brief introduction. First, this tool reads prices, and prices are backward-looking. There will be periods when correlations break abruptly, and patterns fail without warning; no backward-looking, price-based tool will see those moments coming. That is why this is one instrument among several. Our risk management also runs on fundamental, valuation-based, and assumption-based tools that have nothing to do with recent price behavior. Second, better measurement does not guarantee better outcomes. It improves the odds of asking the right question; markets retain the right to change the answer.

What this work does change is what we can build. A fund that does not need a benchmark to understand its own risk does not need a benchmark to design products either. Ideas can be sized and combined on their merits rather than forced into industry-standard baskets. That freedom already shapes our global innovation fund, and it will shape the funds that follow, including the one we are in the process of launching called GenInnov Epicenters. Do reach out if you want to know more about our next attempt at a genre-busting product.

We ask every company we own to show us its innovation. This is some of ours.