Big Ideas

Blockchain is a decentralized, secure digital ledger that automatically records transactions across many computers. Each block of data contains a list of transactions, with blocks linked to form a chain. Cryptography ensures integrity and immutability of block data. Could broad use of blockchain technology make financial markets better? In their May 2025 paper entitled “Crypto and the Evolution of the Capital Markets”, Tuongvy Le and Austin Campbell explore how use of blockchain to trade conventional assets could mitigate or eliminate many of the risks that drive inefficiencies, opaqueness, embedded rent-seeking (frictions), conflicts of interest and concentrations of data and permissions within current market infrastructure. Drawing on lessons from the historical evolution of markets, they conclude that:

Keep Reading

How can investors avoid out-of-sample factor investing strategy failures driven by use of non-causal research methods? In their May 2025 paper entitled “A Protocol for Causal Factor Investing”, Marcos Lopez de Prado and Vincent Zoonekynd introduce the concept of the factor mirage, a factor model that appears statistically valid but is causally mis-specified. They then provide a practical 7-step protocol for causal factor investing that exploits advances in econometrics and machine learning. Based on theoretical analysis, they find that:

• Collider and confounder biases embedded in standard regressions can generate misleading inferences, poor out-of-sample performance and bad investment decisions.
  • Colliders are variables that are causally downstream for both an independent variable and the dependent variable.
    • They often change the sign of coefficients, inducing investors to buy (sell) when they should sell (buy).
    • They are particularly dangerous for standard two-pass or three-pass regressions.
    • Standard evaluation metrics such as adjusted R-squared and t-statistic may reward misspecification and penalize simplicity even when extra variables introduce collider bias.
  • Confounders are uncontrolled variables that cause both an independent variable and the dependent variable. For example, if leverage influences both book-to-market and returns, and is ignored, the estimated role of book-to-market may be biased in magnitude and sign.
• The seven steps for proper causal factor investing are:

• 1. Identify variables potentially associated with future returns.
  2. Create a causal graph that captures variable interdependencies.
  3. Apply do-calculus to the graph to select appropriate control variables.
  4. Use double machine learning to identify causal effects. 
  5. Construct a portfolio that reflects resulting return and risk forecasts.
  6. Backtest portfolio performance via: (a) walk-forward testing; (b) resampling; and, (c) Monte Carlo simulation.
  7. Correct for the elevated risk of false discoveries due to multiple hypothesis testing through p-value and Sharpe ratio adjustments

In summary, investors can produce robust factor investing strategies by using causal reasoning in strategy development.

The paper includes a due diligence questionnaire to assess whether a factor investing scheme is based on causal reasoning.

Cautions regarding findings include:

• The study does not demonstrate/quantify performance improvements in factor investing strategies attributable to use of causal reasoning.
• The methodology described is beyond the reach of most investors, who would bear administrative costs and management fees for delegating to an expert fund manager. Use of the methodology may be costly.

For related research, see results of this search. See also the Compendium of Live ETF Factor/Niche Premium Capture Tests.

Does the evolution of empirical asset pricing point inevitably to machine learning methods? In his February 2025 paper entitled “From Econometrics to Machine Learning: Transforming Empirical Asset Pricing”, Chuan Shi summarizes the transition from traditional methods to machine learning in empirical asset pricing. He traces the historical development of traditional asset pricing models and their roles as benchmarks for decades of research. He compares the strengths and weaknesses of traditional methods and machine learning, explaining why the latter is well-suited to address challenges of the big data era. Finally, he introduces an approach based on the stochastic discount factor (SDF), melding the simplicity of traditional models and the flexibility/predictive power of machine learning. Based on the body of research on asset pricing, he concludes that: Keep Reading

“Compendium of Live ETF Factor/Niche Premium Capture Tests” summarizes results for its eponymous title. Here we add a live test of the short-term reversal effect among U.S. stocks. Specifically, we examine the performance of Vesper U.S. Large Cap Short-Term Reversal Strategy ETF (UTRN), designed to track the performance of a portfolio of 25 of the 500 largest U.S.-listed stocks most likely benefit from the short-term reversal effect. We use SPDR S&P 500 ETF Trust (SPY) as the benchmark. We focus on monthly return statistics, along with compound annual growth rates (CAGR) and maximum drawdowns (MaxDD). Using monthly total returns for UTRN and SPY during September 2018 (UTRN inception) through February 2025, we find that:

Keep Reading

Every review and analysis, including updates of items in “Compendium of Live ETF Factor/Niche Premium Capture Tests”, brings to mind…

• Wherefrom data snooping bias?
  • Data that involve considerable randomness (luck to be discovered).
  • Brute force experimentation with samples/sample periods, model formulas and model parameter values (finding the luck).
  • Reusing and tweaking models and parameter values previously snooped by others (compounding the luck).
  • A community of thousands (millions?) in loose collaboration analyzing the same samples with uncounted model variations (socializing/amplifying luck discovery).
• Implementation frictions are hard to model and often excluded.
• Markets adapt.

Are correlations and R-squared statistics sufficient to discover reliable connections between financial/economic variables and future asset returns? In their February 2025 paper entitled “Causal Factor Analysis is a Necessary Condition for Investment Efficiency”, Marcos Lopez de Prado, Alexander Lipton and Vincent Zoonekynd assess the consequences of factor model misspecification for portfolio optimization (maximizing expected return per unit of risk). They define risk conventionally as: (1) standard deviation of expected portfolio returns; and (2) portfolio exposure to factors, such as size, value, growth, momentum, quality, yield, low volatility, carry, liquidity or macroeconomic variables. Based on theoretical derivations, they conclude that: Keep Reading

Are the human choices in studies that apply machine learning models to forecast stock returns critical to findings? In other words, is there a confounding machine learning design choices zoo? In their November 2024 paper entitled “Design Choices, Machine Learning, and the Cross-section of Stock Returns”, Minghui Chen, Matthias Hanauer and Tobias Kalsbach analyze effects of varying seven key machine learning design choices: (1) machine learning model used, (2) target variable/evaluation metric, (3) target variable transformation (continuous or discrete dummy), (4) whether to use anomaly inputs from pre-publication subperiods or not, (5) whether to compress correlated features, (6) whether to sue a rolling or expanding training window and (7) whether to include micro stocks in the training sample. They examine all possible combinations of these choices, resulting in 1,056 machine learning models. For each machine learning model each month, they:

1. Rank stocks on each of 207 potential return predictors and map rankings into [-1, 1] intervals. In case of missing inputs, they set the ranking value to 0.
2. Apply rankings to predict a next-month target variable (return in excess of the risk-free rate, market-adjusted return or 1-factor model risk-adjusted return) for each stock with market capitalization above a 20% NYSE threshold during January 1987 through December 2021.
3. Reform a hedge portfolio that is long (short) the value-weighted tenth, or decile, of stocks with the highest (lowest) predicted target variable and compute next-month portfolio return.

Using monthly data as available for all listed U.S. common stocks during January 1957 through December 2021, they find that: Keep Reading

What topics are hot and cold in finance research? In their October 2024 paper entitled “Tracing the Evolution of Finance Research: A Topic Modeling Analysis of AJG-Ranked Journals”, Yang Su, Brian Lucey and Ashish Jha provide an overview of academic finance research trends since 2000. They identify the hottest and coldest topics based on publication volume, citation counts and outlier analysis. Using the content and citations for 78,822 articles published across 110 finance journals during 2000 through 2022, they find that:

Keep Reading

How can schemers use statistics to fool investors? In the October 2024 revision of their paper entitled “The Art of Financial Illusion: How to Use Martingale Betting Systems to Fool People”, Carlo Zarattini and Andrew Aziz illustrate use of a Martingale betting system to shape the short-term profitability of trading strategies. This system involves increasing the bet (or trade size) after every loss to recover losses and even yield a profit. Specifically, they run 10,000 trials each for three strategies trading Invesco QQQ Trust (QQQ) daily during 2022, all initially capitalized at $1,000:

1. Base – randomly initiate a 100% long or 100% short position at a random time during regular trading hours with 1:1 leverage and a stop-gain and a stop-loss both $0.20 from the entry price. When no stop triggers, close the position at 4:00PM.
2. Martingale – same as base, but double the leverage after each loss and restore it to 1:1 after a win.
3. Martingale + Target Cumulative Profit – same as base but vary the leverage (in terms of number of shares traded) to target a constant cumulative profit of $0.79 per trading day. In other words, the target profit increases by $0.79 every trading day.

They assume a commission rate of $0.0005 per share. For the second and third strategies, they limit leverage to 500:1. Using intraday prices for QQQ from the end of December 2021 through the end of December 2022, they find that:

Keep Reading

The market view of Bitcoin has increasingly shifted from a potentially useful currency to an investment asset with no yield but potentially high capital gain. What are the implications of its success in the latter role? In their October 2024 paper entitled “The Distributional Consequences of Bitcoin”, Ulrich Bindseil and Jürgen Schaaf model a scenario in which the price of Bitcoin rises for the foreseeable future due to persistent collective belief that its price will continue to rise. Modeling assumptions are:

• All Bitcoin has been mined, such that the supply is constant.
• Bitcoin has no impact on the capacity of the economy to produce goods and services because it has no economic value.
• Success stories of early adopters sustain a steady increase in demand from latecomers, satisfied by early adopter selling. With a fixed supply, price depends exclusively on (rising) demand.
• Latecomers finance Bitcoin purchases by reducing consumption and liquidating real assets (which are bought by early adopters).
• Bitcoin wealth stimulates higher consumption by holders, balanced by the lower consumption of others because Bitcoin does not increase economic activity (ignoring for simplicity the possibility of reduction in other investments). In other words, Bitcoin is a zero sum game.
• Everyone eventually buys some Bitcoin (people never holding Bitcoin would fare worse than latecomers).

Based on market experience with Bitcoin and their model, they conclude that: Keep Reading