Louis Scott

Why do portfolios that look diversified on paper often fail during stress? In our latest Journal of Portfolio Management article, “Reframing Diversification as a Network Problem: Graph Neural Networks for Portfolio Management,” we show how network representations and graph neural networks can uncover structural linkages that are invisible in standard covariance frameworks, and how they can be integrated into portfolio construction, stress testing, and risk management. As markets become more interconnected, portfolio theory must evolve from covariance to connectivity. We hope this framework provides better intuition about how shocks propagate through markets. Che Guan AllianceBernstein Portfolio Management Research https://lnkd.in/e6tHpWtq

162

3 Comments

Quantum computing just gave bond trading a 34% upgrade! HSBC recently disclosed a breakthrough in the use of quantum computing for algorithmic bond trading. They ran a pilot using IBM’s "Heron" processor to predict the likelihood that a bond trade would be filled at a quoted price, achieving a 34% improvement in prediction accuracy. HSBC’s head of technology called it a “Sputnik moment”, signaling the start of a competitive quantum race among the world’s top financial firms. Finance is one of the toughest testing grounds for computing power. If quantum works here, it could transform pharma, AI, energy, materials science, and more! #Investing #Technology

12

2 Comments

https://lnkd.in/dYcDHCQp https://lnkd.in/dBFgbJHB The US Treasury curve reacted to the CPI report yesterday with quite a bit of volatility, but not much decisive movement either way in yields. By EOD, we saw steepening in the curve, with 2 year yields down 4 bps, and 30 year yields up 4 bps. In the belly of curve, things were flat. When the Fed starts cutting, it's rare it stops with 1 cut. The first cut sets the expectation of a cutting cycle. We see that expectation clearly in the Fed Funds, where the highest probabilities show a cut in Sept, a cut in Oct, and a final cut in Dec, which counts 75 bps of cuts into year end. Currently, 2 year treasury yields are about 3.7%, which would put them exactly at the upper end of the target Fed funds rate if this plays out. It is important to note that the Near-Term Forward Rate Spread (3m vs. 3m expected 18 months from now) is inverted by about 50 bps currently. This is down from an inversion of nearly 150 bps last year. Barring any move higher in inflation, 3 rate cuts should finally un-invert this curve. Today, the long end of the curve is reacting, and long rates are going down. The front end is also coming down, in-line with long end, based on our volatility weighted exposures to each point on the curve. To be exact, we are SHORT 1/2 delta in ZT, and SHORT 1/2 delta in ZF. The combined p&l of that short position EXACTLY offsets a 1 delta long position in UB (exposure to 30 year rate). As we get a level decline (where the 3 components of the curve are level, slope, and curve) in the treasury curve, we are seeing a DECLINE in WTI Crude Oil prices, in-line with the QFS position of SHORT 1 delta. This is a well known trade in markets when the driver of CL prices is expectations of economic growth/contraction, as measured by the yield curve. I've attached the notable forecast plots for both UB and CL here.

1

Applying Transformers to Financial Time Series In the previous blog post, we gave a very simple example of how traders can use self-attention transformers as a feature selection method: in this case, to select which previous returns of a stock to use for predictions or optimizations. To be precise, the transformer assigns weights on the different transformed features for downstream applications. In this post, we will discuss how traders can incorporate different feature series from this stock while adding a sense of time. ... Read more at https://lnkd.in/g8NRYAXY

213

5 Comments

Regime models are a critical component of time series forecasting with non-stationary real-world problems. New work by Rebecca Killick and co-authors Thomas Grundy and Ivan Svetunkov tackles this in a novel and I believe very important manner. They are applying change point detection on the forecast errors of *any* predictive model to signal when that model may have passed through a regime change in the underlying inputs and may need to be refit, reevaluated, or discarded. I feel like this is going to be one of those very watershed papers that gets applied all over the place. Use the best predictive model you can, and monitor that model's output reliability with a neutral, transferrable change point model. Highly recommended open access read, with open source code!

36

4 Comments

Structural Regime Dependence of Active Managers As anyone who has ever looked at a SPIVA report knows, the average active manager tends to underperform the market after fees over long periods of time. In this case the benchmark was the S&P500. This small sample analysis broadly confirms that. So the question becomes: Under what market circumstances, regimes, or states do active managers perform relatively better and can we begin to understand why? One idea I have been exploring is that active management may be far more dependent on underlying market structure than most investors realise. Not just whether markets are going up or down, but whether they are: broad or narrow, highly correlated or differentiated, concentrated in a handful of mega caps or dispersed across many stocks, dominated by a single market “mode” or supported by multiple independent drivers. Conceptually, we understand that off-market-cap-weight “bets” should have a greater opportunity to pay off when correlations are lower and markets are more differentiated. But does the data actually support that intuition? Using rolling measures such as breadth, dispersion, average correlation, effective rank, and eigen concentration, I started testing whether different active managers appear to perform better or worse under different structural market regimes. The early results are quite interesting. For example: Vanguard US Growth (VWUSX) appeared to perform materially better in forward periods with higher effective rank and lower average correlation, suggesting a more differentiated stock-picking environment. T Rowe Price Blue Chip Growth (TRBCX) showed particularly strong forward alpha during periods where upside participation dominated downside volatility. What is becoming increasingly interesting is that the patterns are not the same across managers. Different managers appear to require different market “topologies” for skill to emerge. That may potentially provide a new way of identifying managers that genuinely behave differently from one another, rather than relying solely on traditional style boxes or factor classifications.

14

4 Comments

The asset price forecasts for the next 2-5 years   https://lnkd.in/eskTU6gP These are the forecasts that I wrote back in spring 2025. All the trends have been and will be intact. I thought to write detailed explanations before I post the forecasts. But I never had the time. Instead of waiting forever, I will post the forecasts now. Note that many of the predictions will go opposite to my long term forecasts since 2011. Whether those long term forecasts are still valid will depend on how America emerges from this changing era. I am pessimistic about it but maybe Americans can get it right eventually this time as well. A.    China will probably try to take over Taiwan (70% chance), along with the most advanced chip producer, by embargoing it before Trump’s term ends. Most likely China will succeed. So any foreign companies with sizable exposure to Chinese market are vulnerable (e.g., AAPL, TSLA). The prices for things like advanced chips and rare earth will shoot up (MP, CRML, METC, AREC). Even if China doesn’t take over Taiwan, the split between China and the US as well as the West will become reality. China’s rise is a result of globalization, yet also incompatible with globalization and is digging the grave for globalization. B.    China’s equity will have a sharp upward rally beyond any previous peaks. Emerging markets will also have a sharp upswing, ending the dominance of the U.S. stocks markets and US dollar from 2011, which was when I predicted the upswing of the U.S. stock markets and the huge underperformance of emerging markets. Although I have been long term bullish on India from 2011, I would prefer Brazil in this upswing (ETF: EWZ). C.    There will be a period of fast inflation. Oil prices (after a dip) and prices of industrial and precious metals will likely go up a lot, along with the stock prices of metal and oil producers. Such producers in emerging markets will benefit particularly, but US producers such as CLF would too. D.   Interest rates will go up over time globally, probably by a lot. E.    USD will depreciate substantially F.    Gold will go above $10,000 a troy ounce. “I have seen the current correction coming and gold price will probably drop from almost 4400 to at least 3500, and gold miners’ stock prices including AEM’s could drop by half before this correction is over.” (I added what’s in the parenthesis on October 23, 2025) One can keep holding AEM for the next couple years though. G.   U.S. housing price in the U.S. will peak this year, but it may not crash as hard as the 2006-2011 period. H.   U.S. will likely have a large stock market bubble, then probably a recession along with a bear market with stock market dropping substantially before or soon after Trump’s term ends.  I.      The stock and housing prices may not drop as much in nominal term due to inflation and monetary delusion. In real terms, they may look much more terrible.  

16

1 Comment

A thesis is only as good as its collision with reality. At the start of this year, I shared my reflections on 2025: the widening gap between institutional inertia and startup agility, the rise of Model Arbitrage, and my conviction that physical infrastructure is just as strategic as the models themselves. Last week in San Francisco, that thesis was pressure-tested. Here are my observations. What I saw confirmed on the ground: • Agility is winning. Model Arbitrage, treating AI models as a commodity rather than a partner, is now the baseline. Companies locked into a single model are falling behind. Those orchestrating across multiple models and deeply integrating proprietary data into customer workflows are building genuine moats. • The physical bottleneck is real. The AI application layer everyone is excited about simply doesn't scale without massive investment in energy and grid modernization. When roughly 75% of the US power grid's transformers are past their engineered lifespan, physical infrastructure isn't a side theme. It's the binding constraint. The True Economic Magnitude: Beyond infrastructure and models, San Francisco brought the sheer scale of this transition into sharp relief. This isn't just a technology shift that affects labor. It's a labor shift enabled by technology. AI isn't just making workers more productive. It's doing the work itself. The addressable market is moving from IT budgets (roughly $400B) to payroll (around $7 trillion in the US alone). When the TAM expands by that magnitude, you start seeing companies hit revenue milestones at speeds that simply didn't happen in the software era. What this means for how I allocate: The gap between top-tier venture firms and the rest is widening. Not gradually, but as a step function. For allocators, spreading capital across a broad set of decent managers is increasingly a losing proposition. I'd rather concentrate in fewer, deeper relationships with the firms that are actually built for this environment. Every major shift rewards those who move before the consensus forms. The consensus is still forming. I'd rather be early and wrong on timing than late and right on the thesis.

60

4 Comments

The latest Freakonomics episode explores a critical shift in the traditionally private investment space: access to private markets is now expanding to include everyday retail investors like you and me. With traditional institutional capital sources for large PE funds drying out, opening the market to retail capital may create a meaningful funding channel. A few implications are worth noting... - Job growth across the industry is likely to manage increased operational and regulatory needs. - Many retirement plans may soon include private markets investments, which brings new implications regarding taxes and liquidity. - Greater transparency and risk management will be essential in ensuring protection for retail investors. 💭Do you think this will supply a meaningful new source of capital for private equity firms across the board, or primarily benefit the mega-funds? 💭Does this shift strengthen the industry, or simply prolong an inevitable industry restructuring? What are your thoughts? Join the conversation in the comments. https://lnkd.in/gC-z4Tdw

2

📢 *Risk aggregation and stochastic dominance for a class of heavy-tailed distributions* published in ASTIN Bulletin - The Journal of the IAA. Authors: Yuyu Chen, Seva Shneer https://lnkd.in/dBip6KRQ This paper introduces a new class of heavy-tailed distributions for which any weighted average of independent and identically distributed random variables is larger than one such random variable in (usual) stochastic order. It shows that many commonly used extremely heavy-tailed (i.e., infinite-mean) distributions, such as the Pareto, Fréchet, and Burr distributions, belong to this class. The established stochastic dominance relation can be further generalized to allow negatively dependent or non-identically distributed random variables. #actuarialscience #insurance #stochasticorder #riskmanagement #extremevaluetheory

15

Markets didn’t sell off - they repriced. Joined Prof G Media with Ed Elson to talk through what’s really driving the recent tech weakness. My take: • This has largely been a multiple compression story - earnings haven’t changed • The market is recalibrating away from long-duration assets toward cash flow and hard assets • The AI trade itself is forcing that shift - it actually requires real capital, not just software multiples Right now, the market doesn’t quite know how to price that transition - and that’s where the volatility is coming from. Good conversation — link below. https://lnkd.in/gAT-4XtZ

23

3 Comments

Man AHL publishes outstanding research on the #CTA space. No question. They just wrote a fascinating analysis of dispersion among CTAs and different #trendfollowing models. Short answer: diversify. (My answer: replicate 😀) My only quibble is that the model numbers seem off by, well, a lot. Over 25 years, the average Sharpe ratio (see below, vertical axis) seems to be around 0.75. The awkward reality is that the actual Sharpe ratio of the SG CTA Trend index is 0.36. Assuming a 12% average vol, 0.4 of "missing" Sharpe ratio is 500 bps a year. 🤯 If I'm reading the cumulative log returns chart correctly, 18 of the twenty models have outperformed actual #hedgefunds; pick the best one (long term, alts markets, naturally) and you made 13x your money over 25 years, a mere 4-5x the index. 🚀 🚀 🚀 🚀 🚀 Given that they have five (!!!) co-authors, it's entirely possible that I'm missing something. If so, someone poke a hole and I'll eat humble pie. 🥧 If not, then the most likely explanation is that true implementation costs are consistently higher than managers publicly estimate. Here's a link to a recent note I wrote that shows how higher than expected implementation costs might be a factor in why #ETFs have been outperforming hedge funds and #mutualfunds: https://lnkd.in/e6XKyQfn Here's a link to Man's paper: https://lnkd.in/eMBWMnRe

116

17 Comments

On alpha estimation traps: "Conventional approaches may oversimplify risk exposures, rely narrowly defined benchmarks, omit relevant factors, and misattribute beta timing to alpha, resulting in biased estimates that conflate genuine skill with statistical artifacts or unaccounted systematic risk." https://lnkd.in/ey7wVhji

8

𝐍𝐞𝐰 𝐆𝐫𝐨𝐮𝐧𝐝 𝐢𝐧 𝐀𝐦𝐞𝐫𝐢𝐜𝐚𝐧 𝐎𝐩𝐭𝐢𝐨𝐧 𝐏𝐫𝐢𝐜𝐢𝐧𝐠 Despite significant advancements in machine learning for derivative pricing, the efficient and accurate valuation of American options has long been a thorny challenge - complex exercise boundaries, tricky near-expiry behavior, and intricate contract features make traditional methods slow and cumbersome. But what if there was a faster, more accurate way? Our latest research ( https://lnkd.in/dwyWQyu2 ) tackles this problem head-on, extending a semi-analytical approach to price American options under: ✅ Pure diffusions including a multi-dimensional case ✅ Jump-diffusions  ✅ Lévy processes By deriving and solving Volterra integral equations under suitable terminal conditions, we explicitly determine the exercise boundary, outperforming traditional finite-difference and Monte Carlo methods. Key breakthroughs:  🔹 Based on results of Prof. Goran Peskir, generalizing the decomposition method (European price + early exercise premium) to multidimensional diffusions , jump-diffusion and Lévy models. 🔹 Extending the Generalized Integral Transform method (GIT) for doing the same things. 🔹 Handling cases without closed-form transition densities by using characteristic functions (e.g., COS method, or NUFFT method by L. Andersen and M. Lake). Numerical results confirm faster, easier, and more robust pricing – that could be helpful for large-scale financial applications. #QuantFinance #OptionPricing #FinancialEngineering

107

2 Comments

Paper "estimates equity tail risk premia by decomposing the squared VIX into four components: tail risk premium, realised tail, variance risk premium, and realised variance." https://lnkd.in/exZTaVvU

21

1 Comment