publications
Peer-reviewed articles and working papers on machine learning in finance, volatility forecasting, reinforcement learning for trading and hedging, and decentralized finance.
Published Papers
2026
-
Alessio BriniIn Proceedings of the 2026 IEEE 4th International Conference on Artificial Intelligence, Blockchain, and Internet of Things (AIBThings 2026), 2026EIP-4844 gave Ethereum rollups a separate fee dimension for data, governed by a multiplicative update rule that targets three “blobs” per block. Whether this target is reached depends on how many independent, self-interested rollups choose to post data, a question that closed-form models address only for a single agent under exogenous prices or in a static equilibrium. We propose a calibrated multi-agent reinforcement learning environment for the blob fee market in which the blob base fee is endogenous, set by the EIP-4844 rule on the aggregated decisions of heterogeneous rollups fit to post-Dencun Ethereum data. After confirming that the environment reproduces the known single-agent analytical optimum, we show that independent learners, with no convergence guarantee, still settle near the target without coordination (3.32 ± 0.01 blobs per block over five seeds). Over a fiftyfold demand range (10× to 500×), supply stays at the target from 10× to 200× and rises above it at higher demand, where the fee can no longer hold it. Posting frequency increases with rollup size, an emergent analog of the size threshold predicted by prior theory. A costless layer-1 calldata outside option pushes the fee down by roughly an order of magnitude while aggregate supply stays at the target, so the fee rule acts as a negative-feedback controller on decentralized supply.
-
Alessio Brini, and Petter N KolmThe Journal of Financial Data Science, 2026Model-free deep reinforcement learning (DRL) offers a flexible framework for sequential decision-making in finance but faces unique challenges from the stochastic, non-stationary nature of financial markets. We examine the proximal policy optimization (PPO) algorithm for dynamic trading with time-varying alpha and price impact. Using a simulated environment with a closed-form optimal policy, we benchmark PPO’s efficiency and accuracy. We demonstrate how methods of bounding and rescaling its continuous action space significantly impact the training and performance of the DRL agent. We find that clipping the action space yields faster in-sample convergence, and rescaling actions to match the actual range of possible trades is essential for unbiased convergence to the optimal solution. In empirical tests on Dow Jones Industrial Average data with bootstrapped alphas, we show that PPO performance improves when signals are stronger and forecasts span multiple horizons. Our findings highlight the importance of domain-specific adaptations, particularly action space engineering and informative state design, when applying DRL to trading.
-
Alessio Brini, Giacomo Toscano, Ginevra Virginia Lombardi, and 1 more authorEnvironmetrics, 2026This paper provides a framework for measuring honey‐production risk that complements standard mean‐based analyses by explicitly targeting downside tail risk. Using hive‐weight data from a large sample of Italian hives over the period 2021–2024, downside tail risk is quantified through the Honey‐at‐Risk (HaR) metric, defined as the quantile of aggregate daily production shocks for groups of hives exposed to homogeneous climatic conditions. The aggregate shock distribution is modeled based on a flexible meta‐distributional framework that combines hive‐specific marginal distributions with a copula for the cluster dependence structure, allowing for improved accuracy in representing heterogeneous marginal behavior and cross‐hive dependence. The implementation of HaR documents substantial spatial heterogeneity, with mountain clusters facing not only higher downside tail risk but also higher variability in the magnitude of the risk measure, compared to plains and hills. Moreover, logistic regressions suggest that the probability of realizing extreme losses, that is, losses smaller than HaR, is negatively associated with the average atmospheric temperature but positively associated with the occurrence of abnormal temperature maxima. Taken together, these results position HaR as an operational complement to elasticity‐based assessments, enabling the identification of high‐risk areas and supporting targeted adaptation measures as well as the design of weather‐indexed insurance and compensation schemes.
-
Luciano Juvinski, Haochen Li, and Alessio BriniBlockchain, 2026Over $3.1 trillion in illicit money moves through the global financial system yearly, and much of it now uses stablecoins, which launderers prefer for their liquidity. Decentralized protocols increasingly hide transaction patterns with zero-knowledge proofs, while centralized stablecoins remain visible. To preserve the ability to convert to fiat, they must maintain an auditable record, which makes them natural points of compliance oversight. Using this visibility, we create an Ethereum dataset of Tether USD (USDT) and USD Coin (USDC) wallet transfers and establish a baseline for behavioral anti-money-laundering (AML) detection. We compare linear models, tree ensembles, deep networks, and graph neural networks. Tree ensembles achieve the best Macro-F1 score, while the graph neural networks lose accuracy as the transaction network fragments. The models separate distinct typologies rather than only flagging suspicion: the fast, dispersed movement of cybercrime wallets is distinguished from the constrained, static footprint of sanctioned or frozen wallets. These results align with the industry shift toward deterministic verification and address the auditability and compliance requirements now forming under regulations such as the EU’s Markets in Crypto-Assets (MiCA) and the U.S. Guiding and Establishing National Innovation for U.S. Stablecoins Act (GENIUS Act), while limiting unjustified asset freezes. A high-precision behavioral classification of suspicious wallets raises the economic cost of financial misconduct and informs compliance practice under emerging stablecoin rules.
2025
-
Qingran Zhou, Eric Liu, and Alessio BriniIn Proceedings of the 6th ACM International Conference on AI in Finance, 2025Uniswap v3’s design fragments liquidity across thousands of separate pools, giving rise to cross-pool bridge swaps that transmit price changes and redistribute value across liquidity providers. To anticipate these events, we perform six-hour-ahead prediction of bridge-swap links by framing cross-pool trades as a dynamic link-prediction task on a time-evolving, directed graph of liquidity pools. We use on-chain data from January 2022 through June 2024 to evaluate our methods. Our comparison includes logistic regression, tree-based ensembles, sequence models, and graph neural networks. The top performer is a two-layer GraphSAGE network with an edge-aware decoder that integrates temporal node and edge features with neighborhood context. Feature and node-importance analysis highlights cross-pool price ratios and low-fee USDC-WETH and WETH-USDT pools as primary drivers of bridge swaps. Our approach delivers interpretable, forward-looking signals for traders, liquidity providers, and protocol designers to anticipate and manage evolving liquidity dynamics in DeFi.
-
Alessio Brini, and Giacomo ToscanoInternational Journal of Forecasting, 2025This paper introduces SpotV2Net, a multivariate intraday spot volatility forecasting model based on a Graph Attention Network architecture. SpotV2Net represents assets as nodes within a graph and includes non-parametric high-frequency Fourier estimates of the spot volatility and co-volatility as node features. Further, it incorporates Fourier estimates of the spot volatility of volatility and co-volatility of volatility as features for node edges, to capture spillover effects. We test the forecasting accuracy of SpotV2Net in an extensive empirical exercise, conducted with the components of the Dow Jones Industrial Average index. The results we obtain suggest that SpotV2Net yields statistically significant gains in forecasting accuracy, for both single-step and multi-step forecasts, compared to a panel heterogenous auto-regressive model and alternative machine-learning models. To interpret the forecasts produced by SpotV2Net, we employ GNNExplainer (Ying et al., 2019), a model-agnostic interpretability tool, and thereby uncover subgraphs that are critical to a node’s predictions.
-
Alessio Brini, and Viviana FanelliSoft Computing, 2025In our work, we model the dynamics of a startup ecosystem through a delayed differential equation that considers the system growth’s driving factors. The work presents a theoretical and numerical analysis to study the equilibrium and the stability of the described time delay model. The former ensures the model’s desirable characteristic and the equilibrium solution’s local stability. The latter serves the purpose of studying global stability by varying the equation parameters. This study can devise helpful insights for policymakers, who take action to foster business incentives in their area of operation, and entrepreneurs, who can observe a quantitative assessment of an information spillover in a given startup ecosystem.
2024
-
Alessio Brini, and Jimmie LenzFinancial Innovation, 2024The paper analyzes the cryptocurrency ecosystem at both the aggregate and individual levels to understand the factors that impact future volatility. The study uses high-frequency panel data from 2020 to 2022 to examine the relationship between several market volatility drivers, such as daily leverage, signed volatility and jumps. Several known autoregressive model specifications are estimated over different market regimes, and results are compared to equity data as a reference benchmark of a more mature asset class. The panel estimations show that the positive market returns at the high-frequency level increase price volatility, contrary to what is expected from the classical financial literature. We attributed this effect to the price dynamics over the last year of the dataset (2022) by repeating the estimation on different time spans. Moreover, the positive signed volatility and negative daily leverage positively impact the cryptocurrencies’ future volatility, unlike what emerges from the same study on a cross-section of stocks. This result signals a structural difference in a nascent cryptocurrency market that has to mature yet. Further individual-level analysis confirms the findings of the panel analysis and highlights that these effects are statistically significant and commonly shared among many components in the selected universe.
-
Alessio Brini, and Jimmie LenzEconomic Modelling, 2024Pricing cryptocurrency options, crucial for risk management and market stabilization, presents unique challenges due to specific underlying dynamics like the inversion of the leverage effect. Classical option pricing models like Black–Scholes and Heston struggle to address these dynamics due to their set of assumptions. This study introduces machine learning models for options pricing, specifically regression-tree methods. A data-driven machine learning model can incorporate high-frequency volatility estimators into the input set to enhance pricing accuracy. By integrating these estimators, machine learning models can capture the complex dynamics of cryptocurrency markets more effectively than classical pricing approaches. The comparative analysis reveals that equity options are easier to price, clearly indicating inefficiencies in the cryptocurrency option market, which confirms the challenges in achieving accurate pricing. Our results highlight the effectiveness of machine learning models in adapting to the unique characteristics of emerging asset classes, suggesting a shift towards more data-oriented pricing methodologies.
- Alessio Brini, Giacomo Domeniconi, and Ali FathiIn Proceedings of the 5th ACM International Conference on AI in Finance, 2024
We consider the problem of hedging a European call option under a discrete rebalancing schedule with trades subject to transaction costs. We formulate this as a stochastic optimal control problem aiming to maximize the hedge P&L with the quadratic variation of the P&L as a penalty term. We solve the optimization numerically, using deep stochastic optimal control and deep reinforcement learning when the market follows either the Black-Scholes or a stochastic volatility model. Furthermore, under the Black-Scholes model, we show that delta hedging is not the optimal hedging strategy when penalized for the quadratic variation. Our results show that data-driven methods outperform traditional delta-hedging strategies when accounting for transaction costs and pathwise variability. We observe how these methods are well-suited for multi-step optimization problems and can effectively balance hedging costs over time.
2023
-
Alessio Brini, Gabriele Tedeschi, and Daniele TantariJournal of Financial Stability, 2023In this paper, we analyze the effect of a policy recommendation on the performance of an artificial interbank market. Financial institutions stipulate lending agreements following a public recommendation and their individual information. The former is modeled by a reinforcement learning optimal policy that maximizes the system’s fitness and gathers information on the economic environment. The policy recommendation directs economic actors to create credit relationships through the optimal choice between a low interest rate or a high liquidity supply. The latter, based on the agents’ balance sheet, allows determining the liquidity supply and interest rate that the banks optimally offer their clients within the market. Thanks to the combination between the public and the private signal, financial institutions create or cut their credit connections over time via a preferential attachment evolving procedure able to generate a dynamic network. Our results show that the emergence of a core-periphery interbank network, combined with a certain level of homogeneity in the size of lenders and borrowers, is essential to ensure the system’s resilience. Moreover, the optimal policy recommendation obtained through reinforcement learning is crucial in mitigating systemic risk.
-
Alessio Brini, and Daniele TantariPhysica A: Statistical Mechanics and its Applications, 2023Classical portfolio optimization often requires forecasting asset returns and their corresponding variances in spite of the low signal-to-noise ratio provided in the financial markets. Modern deep reinforcement learning (DRL) offers a framework for optimizing sequential trader decisions but lacks theoretical guarantees of convergence. On the other hand, the performances on real financial trading problems are strongly affected by the goodness of the signal used to predict returns. To disentangle the effects coming from return unpredictability from those coming from algorithm un-trainability, we investigate the performance of model-free DRL traders in a market environment with different known mean-reverting factors driving the dynamics. When the framework admits an exact dynamic programming solution, we can assess the limits and capabilities of different value-based algorithms to retrieve meaningful trading signals in a data-driven manner. We consider DRL agents that leverage classical strategies to increase their performances and we show that this approach guarantees flexibility, outperforming the benchmark strategy when the price dynamics is misspecified and some original assumptions on the market environment are violated with the presence of extreme events and volatility clustering.
2022
-
Alessio Brini, and Jimmie LenzEconomics Letters, 2022By analyzing a large cross-section of cryptocurrencies, we document the absence of the leverage effect in this market. Unlike the equity market, investors exhibit less panicking behavior and appear indifferent to negative returns in terms of market participation. Moreover, the negative asymmetric effect is reverted for some cryptocurrencies in our dataset, showing the investors’ fear of missing out. Our results are robust over different leverage effect models and historical time windows.
Working Papers
2026
- Célestin Coquidé, Alessio Brini, and Natkamon TovanicharXiv preprint arXiv:2610.23466, 2026
Decentralized Finance (DeFi) transactions involve complex interactions among accounts, smart contracts, and tokens. We investigate whether network topology can identify transaction methods that describe DeFi operations. We extract directed graphlet features to classify the 30 most frequent methods. Our results show that graphlet features complement event log counts in improving classification performance. Using SHapley Additive exPlanations (SHAP), we further derive model explanations that reveal method-specific graphlet structural patterns.
- Alessio Brini, and Ekaterina SereginaarXiv preprint arXiv:2601.16274, 2026
We propose the Mixed-Panels-Transformer Encoder (MPTE), a framework for estimating factor models in panels with mixed frequencies and nonlinear signals. Classical factor models rely on linear signal extraction and homogeneous sampling frequencies, limiting their use when variables arrive at different frequencies. MPTE instead uses Transformer-style attention to construct context-aware signals, replacing fixed linear combinations with adaptive reweighting. We extend principal component analysis to accommodate general temporal and cross-sectional attention operators, so the model learns to aggregate information across frequencies without manual alignment. Under linear activations, we establish consistency and asymptotic normality of factor and loading estimators, show that the framework nests classical factor models as a special case, and obtain efficiency gains through transfer learning across auxiliary panels. A Transformer architecture handles the nonlinear case, which we assess through simulations and an empirical application. In simulations, MPTE outperforms linear benchmarks under nonlinear designs. On 13 quarterly U.S. macroeconomic targets drawn from 48 monthly and quarterly FRED series, it remains competitive with established benchmarks. By averaging learned attention across variables and time, we recover target-specific variable importance and lag relevance, and ablations quantify the contribution of each model component.
- Alessio BriniarXiv preprint arXiv:2607.05291, 2026
We ask whether pretrained time series foundation models (TSFMs) improve on established econometric benchmarks for forecasting realized volatility. Using the VOLARE dataset, we conduct the first systematic comparison of nine zero-shot TSFMs against eight econometric specifications, including the Heterogeneous Autoregressive (HAR) family, across 50 assets in equities, foreign exchange, and futures, and three forecast horizons, with formal pairwise and multi-model forecast-comparison tests. Foundation models do not deliver a uniform gain. Pooled losses favor them, but the advantage is concentrated in a few outlier assets; averaging each asset’s loss ratio to a well-specified Log-HAR benchmark, so that no single asset dominates, only one small model, Tiny Time Mixers (TTM), beats the benchmark at every horizon, and by a narrow margin. The other foundation models do not improve on Log-HAR, and the econometric benchmarks remain competitive throughout. A Mincer–Zarnowitz recalibration, which removes level and scale bias from every forecast, shows that much of the short-horizon advantage reflects better-scaled forecasts rather than better prediction of volatility dynamics, and only at the monthly horizon does a genuine informational gain remain. Because this edge is thin and even TTM is not best on every asset, a simple equal-weight average of TTM and Log-HAR matches the best single model and enters the Model Confidence Set for 98 to 100% of assets, more often than either component alone, so a forecaster need not identify the best model for each asset in advance. Our most durable finding is that performance varies so much across foundation-model architectures that choosing the right architecture matters more than the broader choice between foundation and econometric models.
2025
- Haonan Xu, and Alessio BriniarXiv preprint arXiv:2501.07508, 2025
This paper applies deep reinforcement learning (DRL) to optimize liquidity provisioning in Uniswap v3, a decentralized finance (DeFi) protocol implementing an automated market maker (AMM) model with concentrated liquidity. We model the liquidity provision task as a Markov Decision Process (MDP) and train an active liquidity provider (LP) agent using the Proximal Policy Optimization (PPO) algorithm. The agent dynamically adjusts liquidity positions by using information about price dynamics to balance fee maximization and impermanent loss mitigation. We use a rolling window approach for training and testing, reflecting realistic market conditions and regime shifts. This study compares the data-driven performance of the DRL-based strategy against common heuristics adopted by small retail LP actors that do not systematically modify their liquidity positions. By promoting more efficient liquidity management, this work aims to make DeFi markets more accessible and inclusive for a broader range of participants. Through a data-driven approach to liquidity management, this work seeks to contribute to the ongoing development of more efficient and user-friendly DeFi markets.
- Alessio Brini, David A Hsieh, Patrick Kuiper, and 2 more authorsarXiv preprint arXiv:2506.17511, 2025
The key objective of this paper is to develop an empirical model for pricing SPX options that can be simulated over future paths of the SPX. To accomplish this, we formulate and rigorously evaluate several statistical models, including neural network, random forest, and linear regression. These models use the observed characteristics of the options as inputs – their price, moneyness and time-to-maturity, as well as a small set of external inputs, such as the SPX and its past history, dividend yield, and the risk-free rate. Model evaluation is performed on historical options data, spanning 30 years of daily observations. Significant effort is given to understanding the data and ensuring explainability for the neural network. A neural network model with two hidden layers and four neurons per layer, trained with minimal hyperparameter tuning, performs well against the theoretical Black-Scholes-Merton model for European options, as well as two other empirical models based on the random forest and the linear regression. It delivers arbitrage-free option prices without requiring these conditions to be imposed.
2023
- Alessio Brini, Elisa Giovannini, and Elia SmaniottoarXiv preprint arXiv:2304.01215, 2023
The beekeeping sector has experienced significant production fluctuations in recent years, largely due to increasingly frequent adverse weather events linked to climate change. These events can severely affect the environment, reducing its suitability for bee activity. We conduct a forecasting analysis of honey production across Italy using a range of machine learning models, with a particular focus on weather-related variables as key predictors. Our analysis relies on a dataset collected in 2022, which combines hive-level observations with detailed weather data. We train and compare several linear and nonlinear models, evaluating both their predictive accuracy and interpretability. By examining model explanations, we identify the main drivers of honey production. We also ensemble models from different families to assess whether combining predictions improves forecast accuracy. These insights support beekeepers in managing production risks and may inform the development of insurance products against unexpected losses due to poor harvests.
2022
- Crypto dynamics during market downturns: Another dotcom boom-bust cycle?Alessio Brini, Jimmie Lenz, and Emma RasielDigital Asset Research & Engineering Collaborative, Duke University, 2022
-
Alessio Brini, and Jimmie LenzAvailable at SSRN, 2022An exchange traded fund (ETF) is a financial product that attempts to match the performance of a benchmark, which might be an index like the S&P 500, or a particular commodity. A key characteristic of the ETFs is that they can be traded on stock exchanges, just as individual equities are traded. This is particularly valuable to retail investors, for whom obtaining exposure to equity indices or commodities is otherwise only available through mutual funds. ETFs provide cheaper and easier access to a huge range of assets and indices, and hence have become very popular. Recent additions to the ETF ecosystem have included products that attempt to track the price of Bitcoin (BTC), the most valuable cryptocurrency among digital assets. ProShares Bitcoin Strategy ETF (BITO) began trading on October 18th last year, followed by Valkyrie (BTF) and VanEck (XBTF). Managers of these products use the futures markets to track the price of Bitcoin, yet there are no ETFs that tie directly to Bitcoin’s spot price. The use of futures to track BTC may introduce a difference between the return of the ETF and that of its underlying benchmark. The standard deviation of this difference – the so-called “tracking error” – is the typical measure of performance for an ETF relative to its index and can be of considerable magnitude, which is troubling, since these ETFs have already attracted billions in investment dollars. Furthermore, tracking error alone may be insufficient as a measure of ETF performance. The less commonly used “tracking difference” metric provides additional insight into BTC-benchmarked ETF that should also be taken into consideration. In this article, we disentangle the various components of tracking error, and explain the use of the additional tracking difference metric to provide an enhanced framework for evaluating the Bitcoin ETFs performance.
