A collection of our white papers and peer-reviewed research articles related to macro / multi-asset investor behavior, hedging, risk regimes, liquidity risk, private assets, portfolio construction, and more.
By Megan Czasonis, Yin Li, Huili Song, and David Turkington
Our innovative "interrogation" method detects unreliable machine learning predictions in advance, overcoming limitations of the traditional cross-validation method.
We introduce a new method called "interrogation" to warn when a machine learning model has underfit or overfit a data sample, offering a more efficient alternative to traditional cross-validation. Unlike cross-validation, which can be cumbersome and computationally expensive, interrogation evaluates models trained on all available data by breaking down their prediction logic into linear, nonlinear, pairwise, and high-order interaction components. This method successfully identified near-optimal stopping times for training neural networks without using validation samples, boosting confidence that models are well-calibrated and can perform reliably on new data. Interrogation is model-agnostic, providing transparency and reliability even for black-box models.
By Alexander Cheema-Fox, Megan Czasonis, Piyush Kontu and George Serafeim
We explore the world’s first set of financial accounting data on firms’ sustainable activities.
Though sustainable investing has grown in popularity over the past decade, measuring sustainability remains a key challenge. Investors often rely on environmental criteria—such as analyst ratings and carbon emissions—that are insufficient or rely on qualitative analysis. However, for the first time, with the advent of the EU’s Taxonomy for Sustainable Activities, investors have access to financial accounting data that follows standardized and transparent criteria for quantifying the percentage of a firm’s revenues and expenditures that align with sustainable activities. In a recent paper, we explore this novel dataset for a cross-section of large European firms, documenting patterns and analysing how firms’ aligned activities relate to fundamentals and environment ratings. We find that the EU Taxonomy data provide information that is distinct from existing sources and offers insights that can help investors and regulators, alike.
By Megan Czasonis, Mark Kritzman, and David Turkington
We show that relevance-based prediction offers an elegant solution to the problem of incomplete data, preserving valuable information and enhancing prediction reliability in a way that is not possible using traditional models.
When setting out to form data-driven predictions, it’s common to encounter incomplete information, such as a time series with shorter history lengths or observations with missing data. Traditional methods for addressing this challenge either discard valuable data or manufacture replacements based on limiting assumptions, leading to unreliable results. We propose a novel technique called Relevance-Based Prediction (RBP), which elegantly navigates the pitfalls of missing data by retaining more information and accounting for the relative importance of observations for which only partial data is available. We show that RBP offers an elegant solution to the problem of incomplete data, preserving valuable information and enhancing prediction reliability in a way that is not possible using traditional models.
By Megan Czasonis, Mark Kritzman and David Turkington
We show how our method of relevance-based prediction implements similar logic to a highly complex machine learning model, but relevance is extremely transparent.
What is the best way to form predictions from a data sample? This is a big question, but at its core lies a fundamental tension between explaining the past and anticipating the future. Predictions can fail by paying too little attention to the past (underfitting) or by paying too much attention (overfitting). High-complexity machine learning models address this problem by recombining past information in thousands (or millions) of exotic ways to map out generalized rules for any situation. An alternative method, called relevance-based prediction, considers each situation one at a time, and extracts the past data that is most useful for that task. We show that there is a deep connection between the two approaches, but only relevance maintains the transparency that makes it easy to explain precisely how each past experience informs a prediction.
By Alexander Swade, Matthias X. Hanauer, Harald Lohre, and David Blitz
The authors propose a straightforward yet effective method to identify the factors that capture most of the available "alpha".
Since the introduction of the Capital Asset Pricing Model (CAPM), researchers have been on a quest to find the most important factors, leading to a crowded "factor zoo." Despite the variety of these factors, academic models suggested for years that most of them can be boiled down to just four to six key ones. Recent publications in this area have overcome one key challenge in sorting through this factor zoo: Not only did they reconstruct a majority of factors from the literature and publish them in open-source databases but also addressed the so-called replication crisis in finance with this approach. The authors propose a straightforward yet effective method to identify the factors that capture most of the available "alpha". These findings help investors navigate the complex factor zoo by pinpointing strong alpha contributors and comprehensive models. It also highlights the ongoing need for innovation and adjustments in models.
By Megan Czasonis, Mark Kritzman, and David Turkington
As the race to design sophisticated data analytics continues, we show why relevance-based prediction offers an ideal way to measure the importance of an input variable to a prediction.
T-statistics act as a hallmark for rigor by pinpointing the effect of a single variable and distinguishing signal from noise. However, they have significant limitations: (1) t-stats do not capture ‘shared’ information, (2) t-stats are not prediction-specific, and (3) t-stats only consider linear relationships. In a recent paper, we introduce an alternative method, called Relevance-Based Importance (RBI), which measures the importance of every variable to the reliability of every individual prediction. RBI recognizes that it is almost never the case that a variable is always important, or that it is never important. Rather, it's more likely that variables are sometimes important, depending on the circumstance. We show that RBI brings the virtues of t-statistics but also adapts to each unique situation, making it robust to complexities where t-stats fall short.
By Megan Czasonis, Huili Song, and David Turkington
We show that LLMs can effectively extrapolate from disparate domains of knowledge to reason through economic relationships, and that this may have advantages over narrower statistical models.
Fundamentally, large language models (LLMs) and numerical models both learn patterns in training data. However, while traditional models rely on narrowly curated datasets, LLMs can extrapolate patterns across disparate domains of knowledge. In new research, we explore whether this ability is valuable for predicting economic outcomes. First, we ask LLMs to infer economic growth based on hypothetical conditions of other economic variables. We then use our Model Fingerprint framework to interpret how they use linear, nonlinear, and conditional logic to understand economic linkages. We find that their reasoning is intuitive, and it differs meaningfully from the reasoning of statistical models. We also compare the accuracy of the models’ reasoning using historical data and find that the LLMs infer growth outcomes more reliably than the statistical models. These results suggest that LLMs can effectively reason through economic relationships and that cross-domain extrapolation may add value above explicit statistical analysis.
By William Kinlaw, Mark Kritzman, and David Turkington
Conventional statistics hide important realities that investors need to know.
The correlation coefficient often fails to capture what really matters to investors. There are two reasons for this. First, investors often measure correlations using monthly data and assume that they also hold over one-year, five-year or ten-year periods. Unfortunately, in the real world, they often don’t. The second reason has to do with a fundamental misconception about diversification. The fact is, investors don’t always want it. Sure, they want it on the downside, in order to offset the poor performance of one or more assets. But on the upside they prefer all assets to rise in unison, which is the opposite of diversification. Put differently, they’d be happy to place their eggs, conveniently, in a single basket provided nobody steals it. Our research shows that correlations can vary through time based on a range of conditions including the level of interest rates, the degree of turbulence in financial markets, and the performance of major equity markets. Overall, our findings challenge the notion that returns evolve as a simple “random walk,” a critical pre-condition without which we must interpret the correlation coefficient distrustfully. To address these issues, we introduce the notion of co-occurrence and offer a new perspective on how investors should diversify portfolios.
By Megan Czasonis, Mark Kritzman, and David Turkington
We propose a new currency hedging technique called full-scale hedging, which accounts for the complexities of diversification.
Diversification is nuanced and summary statistics, such as correlation, fail to capture complexities that lie below the surface. For investors, these complexities matter—accounting for them can make the difference between an effective, or ineffective, hedging strategy. In the case of currencies, investors often determine risk-minimizing hedge ratios based on the portfolio’s betas to those currencies or with mean-variance optimization. In both cases, the optimal solution depends crucially on the correlation between the currencies and assets in the portfolio. But correlation is an unreliable estimate of the diversification investors actually care about: the co-occurrences of the cumulative returns of the portfolio and currencies over the investment horizon. We propose a new currency hedging technique called full-scale hedging, which addresses these challenges by considering the full distribution of co-occurrences between currencies and the portfolio.
By Alexander Cheema-Fox and Robin Greenwood
Using a uniquely deep proprietary dataset, we detail how global investors across regions and asset classes hedge their currency risk, stick to their hedges, and adjust their hedging targets over time.
Currency risk is a key component of global investor returns, but different categories of investor approach these exposures differently. Using State Street’s proprietary custodial data, we have a uniquely precise view into how investors actually choose to hedge and how that varies over time, by asset class, and across different investor domiciles. We introduce a new quantity, the “dynamic hedge ratio,” to capture how investors adjust their hedge ratios and rebalance their currency risk over time. We find that US investors hedge less than others, that equity investors hedge less than fixed-income investors, and that investors tend to stick to target hedge ratios. Moreover, we find that average hedge ratios vary through time with currency, equity, and bond factors, yet exhibit a post GFC shift towards higher hedge ratios that cannot be explained by these factors.