nep-cmp New Economics Papers
on Computational Economics
Issue of 2023‒09‒18
twenty-one papers chosen by
Stan Miles, Thompson Rivers University

  1. A Comparison of Neural Networks and Bayesian MCMC for the Heston Model Estimation (Forget Statistics - Machine Learning is Sufficient!) By Jiří Witzany; Milan Fičura
  2. Munging the Ghosts in the Machine: Coded Bias and the Craft of Wrangling Archival Data By Yung, Vincent; Colyvas, Jeannette
  3. Company Similarity using Large Language Models By Dimitrios Vamvourellis; M\'at\'e Toth; Snigdha Bhagat; Dhruv Desai; Dhagash Mehta; Stefano Pasquali
  4. IMM: An Imitative Reinforcement Learning Approach with Predictive Representation Learning for Automatic Market Making By Hui Niu; Siyuan Li; Jiahao Zheng; Zhouchi Lin; Jian Li; Jian Guo; Bo An
  5. Variations on the Reinforcement Learning performance of Blackjack By Avish Buramdoyal; Tim Gebbie
  6. Anomaly Detection in Global Financial Markets with Graph Neural Networks and Nonextensive Entropy By Kleyton da Costa
  7. Machine Forecast Disagreement By Turan G. Bali; Bryan T. Kelly; Mathis Mörke; Jamil Rahman
  8. Nowcasting trade in value added indicators By Annabelle Mourougane; Polina Knutsson; Rodrigo Pazos; Julia Schmidt; Francesco Palermo
  9. Emerging Frontiers: Exploring the Impact of Generative AI Platforms on University Quantitative Finance Examinations By Rama K. Malladi
  10. Artificial Intelligence and Scientific Discovery: A Model of Prioritized Search By Ajay K. Agrawal; John McHale; Alexander Oettl
  11. Meta-Analysis of Social Science Research: A Practitioner´s Guide By Zuzana Irsova; Hristos Doucouliagos; Tomas Havranek; T. D. Stanley
  12. A New Approach to Overcoming Zero Trade in Gravity Models to Avoid Indefinite Values in Linear Logarithmic Equations and Parameter Verification Using Machine Learning By Mikrajuddin Abdullah
  13. The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market By Xiang Hui; Oren Reshef; Luofeng Zhou
  14. Inventor Gender and Patent Undercitation: Evidence from Causal Text Estimation By Yael Hochberg; Ali Kakhbod; Peiyao Li; Kunal Sachdeva
  15. How uncertainty shapes herding in the corporate use of artificial intelligence technology By Nicolas Ameye; Jacques Bughin; Nicolas van Zeebroeck
  16. The FOMC versus the Staff: Do Policymakers Add Value in Their Tales? By Ilias Filippou; James Mitchell; My T. Nguyen
  17. Regulating Artificial Intelligence in the EU, United States and China - Implications for energy systems By Fabian Heymann; Konstantinos Parginos; Ali Hariri; Gabriele Franco
  18. Multi-level Fusion in Deep Convolutional Networks for Enhanced Image Analysis By Mathias, Lea
  19. AI-Assisted Investigation of On-Chain Parameters: Risky Cryptocurrencies and Price Factors By Abdulrezzak Zekiye; Semih Utku; Fadi Amroush; Oznur Ozkasap
  20. Estimating HANK for Central Banks By Sushant Acharya; William Chen; Marco Del Negro; Keshav Dogra; Aidan Gleich; Shlok Goyal; Donggyu Lee; Ethan Matlin; Reca Sarfati; Sikata Sengupta
  21. Nested Multilevel Monte Carlo with Biased and Antithetic Sampling By Abdul-Lateef Haji-Ali; Jonathan Spence

  1. By: Jiří Witzany; Milan Fičura
    Abstract: The main goal of this paper is to compare the classical MCMC estimation method with a universal Neural Network (NN) approach to estimate unknown parameters of the Heston stochastic volatility model given a series of observable asset returns. The main idea of the NN approach is to generate a large training synthetic dataset with sampled parameter vectors and the return series conditional on the Heston model. The NN can then be trained reverting the input and output, i.e. setting the return series, or rather a set of derived generalized moments as the input features and the parameters as the target. Once the NN has been trained, the estimation of parameters given observed return series becomes very efficient compared to the MCMC algorithm. Our empirical study implements the MCMC estimation algorithm and demonstrates that the trained NN provides more precise and substantially faster estimations of the Heston model parameters. We discuss some other advantages and disadvantages of the two methods, and hypothesize that the universal NN approach can in general give better results compared to the classical statistical estimation methods for a wide class of models.
    Keywords: Heston model, parameter estimation, neural networks, MCMC
    JEL: C45 C63 G13
    Date: 2023–07–11
  2. By: Yung, Vincent (Northwestern University); Colyvas, Jeannette
    Abstract: Data wrangling is typically treated as an obligatory, codified, and ideally automated step in the machine learning (ML) pipeline. In contrast, we suggest that archival data wrangling is a theory-driven process best understood as a practical craft. Drawing on empirical examples from contemporary computational social science, we identify nine core modes of data wrangling, which can be seen as a sequence but are iterative and nonlinear in practice. Moreover, we discuss how data wrangling can address issues of algorithmic bias. While ML has shifted the focus towards architectural engineering, we assert that to properly engage with machine learning is to properly engage with data wrangling.
    Date: 2023–08–18
  3. By: Dimitrios Vamvourellis; M\'at\'e Toth; Snigdha Bhagat; Dhruv Desai; Dhagash Mehta; Stefano Pasquali
    Abstract: Identifying companies with similar profiles is a core task in finance with a wide range of applications in portfolio construction, asset pricing and risk attribution. When a rigorous definition of similarity is lacking, financial analysts usually resort to 'traditional' industry classifications such as Global Industry Classification System (GICS) which assign a unique category to each company at different levels of granularity. Due to their discrete nature, though, GICS classifications do not allow for ranking companies in terms of similarity. In this paper, we explore the ability of pre-trained and finetuned large language models (LLMs) to learn company embeddings based on the business descriptions reported in SEC filings. We show that we can reproduce GICS classifications using the embeddings as features. We also benchmark these embeddings on various machine learning and financial metrics and conclude that the companies that are similar according to the embeddings are also similar in terms of financial performance metrics including return correlation.
    Date: 2023–08
  4. By: Hui Niu; Siyuan Li; Jiahao Zheng; Zhouchi Lin; Jian Li; Jian Guo; Bo An
    Abstract: Market making (MM) has attracted significant attention in financial trading owing to its essential function in ensuring market liquidity. With strong capabilities in sequential decision-making, Reinforcement Learning (RL) technology has achieved remarkable success in quantitative trading. Nonetheless, most existing RL-based MM methods focus on optimizing single-price level strategies which fail at frequent order cancellations and loss of queue priority. Strategies involving multiple price levels align better with actual trading scenarios. However, given the complexity that multi-price level strategies involves a comprehensive trading action space, the challenge of effectively training profitable RL agents for MM persists. Inspired by the efficient workflow of professional human market makers, we propose Imitative Market Maker (IMM), a novel RL framework leveraging both knowledge from suboptimal signal-based experts and direct policy interactions to develop multi-price level MM strategies efficiently. The framework start with introducing effective state and action representations adept at encoding information about multi-price level orders. Furthermore, IMM integrates a representation learning unit capable of capturing both short- and long-term market trends to mitigate adverse selection risk. Subsequently, IMM formulates an expert strategy based on signals and trains the agent through the integration of RL and imitation learning techniques, leading to efficient learning. Extensive experimental results on four real-world market datasets demonstrate that IMM outperforms current RL-based market making strategies in terms of several financial criteria. The findings of the ablation study substantiate the effectiveness of the model components.
    Date: 2023–08
  5. By: Avish Buramdoyal; Tim Gebbie
    Abstract: Blackjack or "21" is a popular card-based game of chance and skill. The objective of the game is to win by obtaining a hand total higher than the dealer's without exceeding 21. The ideal blackjack strategy will maximize financial return in the long run while avoiding gambler's ruin. The stochastic environment and inherent reward structure of blackjack presents an appealing problem to better understand reinforcement learning agents in the presence of environment variations. Here we consider a q-learning solution for optimal play and investigate the rate of learning convergence of the algorithm as a function of deck size. A blackjack simulator allowing for universal blackjack rules is also implemented to demonstrate the extent to which a card counter perfectly using the basic strategy and hi-lo system can bring the house to bankruptcy and how environment variations impact this outcome. The novelty of our work is to place this conceptual understanding of the impact of deck size in the context of learning agent convergence.
    Date: 2023–08
  6. By: Kleyton da Costa
    Abstract: Anomaly detection is a challenging task, particularly in systems with many variables. Anomalies are outliers that statistically differ from the analyzed data and can arise from rare events, malfunctions, or system misuse. This study investigated the ability to detect anomalies in global financial markets through Graph Neural Networks (GNN) considering an uncertainty scenario measured by a nonextensive entropy. The main findings show that the complex structure of highly correlated assets decreases in a crisis, and the number of anomalies is statistically different for nonextensive entropy parameters considering before, during, and after crisis.
    Date: 2023–08
  7. By: Turan G. Bali; Bryan T. Kelly; Mathis Mörke; Jamil Rahman
    Abstract: We propose a statistical model of differences in beliefs in which heterogeneous investors are represented as different machine learning model specifications. Each investor forms return forecasts from their own specific model using data inputs that are available to all investors. We measure disagreement as dispersion in forecasts across investor-models. Our measure aligns with extant measures of disagreement (e.g., analyst forecast dispersion), but is a significantly stronger predictor of future returns. We document a large, significant, and highly robust negative cross-sectional relation between belief disagreement and future returns. A decile spread portfolio that is short stocks with high forecast disagreement and long stocks with low disagreement earns a value-weighted alpha of 15% per year. A range of analyses suggest the alpha is mispricing induced by short-sale costs and limits-to-arbitrage.
    JEL: C15 C4 C45 C58 G1 G10 G17 G4 G40
    Date: 2023–08
  8. By: Annabelle Mourougane; Polina Knutsson; Rodrigo Pazos; Julia Schmidt; Francesco Palermo
    Abstract: Trade in value added (TiVA) indicators are increasingly used to monitor countries’ integration into global supply chains. However, they are published with a significant lag - often two or three years - which reduces their relevance for monitoring recent economic developments. This paper aims to provide more timely insights into the international fragmentation of production by exploring new ways of nowcasting five TiVA indicators for the years 2021 and 2022 covering a panel of 41 economies at the economy-wide level and for 24 industry sectors. The analysis relies on a range of models, including Gradient boosted trees (GBM), and other machine-learning techniques, in a panel setting, uses a wide range of explanatory variables capturing domestic business cycles and global economic developments and corrects for publication lags to produce nowcasts in quasi-real time conditions. Resulting nowcasting algorithms significantly improve compared to the benchmark model and exhibit relatively low prediction errors at a one- and two-year horizon, although model performance varies across countries and sectors.
    Keywords: Global value chains, Machine learning, Nowcasting
    JEL: C4 C53 F17
    Date: 2023–09–06
  9. By: Rama K. Malladi
    Abstract: This study evaluated three Artificial Intelligence (AI) large language model (LLM) enabled platforms - ChatGPT, BARD, and Bing AI - to answer an undergraduate finance exam with 20 quantitative questions across various difficulty levels. ChatGPT scored 30 percent, outperforming Bing AI, which scored 20 percent, while Bard lagged behind with a score of 15 percent. These models faced common challenges, such as inaccurate computations and formula selection. While they are currently insufficient for helping students pass the finance exam, they serve as valuable tools for dedicated learners. Future advancements are expected to overcome these limitations, allowing for improved formula selection and accurate computations and potentially enabling students to score 90 percent or higher.
    Date: 2023–08
  10. By: Ajay K. Agrawal; John McHale; Alexander Oettl
    Abstract: We model a key step in the innovation process, hypothesis generation, as the making of predictions over a vast combinatorial space. Traditionally, scientists and innovators use theory or intuition to guide their search. Increasingly, however, they use artificial intelligence (AI) instead. We model innovation as resulting from sequential search over a combinatorial design space, where the prioritization of costly tests is achieved using a predictive model. We represent the ranked output of the predictive model in the form of a hazard function. We then use discrete survival analysis to obtain the main innovation outcomes of interest – the probability of innovation, expected search duration, and expected profit. We describe conditions under which shifting from the traditional method of hypothesis generation, using theory or intuition, to instead using AI that generates higher fidelity predictions, results in a higher likelihood of successful innovation, shorter search durations, and higher expected profits. We then explore the complementarity between hypothesis generation and hypothesis testing; potential gains from AI may not be realized without significant investment in testing capacity. We discuss the policy implications.
    JEL: O31 O33
    Date: 2023–08
  11. By: Zuzana Irsova (Charles University, Prague & Anglo-American University, Prague); Hristos Doucouliagos (Department of Economics and Deakin Laboratory for the Meta-Analysis of Research. Deakin University, Melbourne, Australia.); Tomas Havranek (Charles University, Prague & Centre for Economic Policy Research, London); T. D. Stanley (4Department of Economics and Deakin Laboratory for the Meta-Analysis of Research, Deakin University, Melbourne, Australia)
    Abstract: This paper provides concise, nontechnical, step-by-step guidelines on how to conduct a modern meta-analysis, especially in social sciences. We treat publication bias, p-hacking, and heterogeneity as phenomena meta-analysts must always confront. To this end, we provide concrete methodological recommendations. Meta-analysis methods have advanced notably over the last few years. Yet many meta-analyses still rely on outdated approaches, some ignoring publication bias and systematic heterogeneity. While limitations persist, recently developed techniques allow robust inference even in the face of formidable problems in the underlying empirical literature. The purpose of this paper is to summarize the state of the art in a way accessible to aspiring meta-analysts in any field. We also discuss how meta-analysts can use advances in artificial intelligence to work more efficiently.
    Keywords: meta-analysis, publication bias, p-hacking, artificial intelligence, model uncertainty
    JEL: C83 H52 I21
    Date: 2023–09
  12. By: Mikrajuddin Abdullah
    Abstract: The presence of a high number of zero flow trades continues to provide a challenge in identifying gravity parameters to explain international trade using the gravity model. Linear regression with a logarithmic linear equation encounters an indefinite value on the logarithmic trade. Although several approaches to solving this problem have been proposed, the majority of them are no longer based on linear regression, making the process of finding solutions more complex. In this work, we suggest a two-step technique for determining the gravity parameters: first, perform linear regression locally to establish a dummy value to substitute trade flow zero, and then estimating the gravity parameters. Iterative techniques are used to determine the optimum parameters. Machine learning is used to test the estimated parameters by analyzing their position in the cluster. We calculated international trade figures for 2004, 2009, 2014, and 2019. We just examine the classic gravity equation and discover that the powers of GDP and distance are in the same cluster and are both worth roughly one. The strategy presented here can be used to solve other problems involving log-linear regression.
    Date: 2023–08
  13. By: Xiang Hui; Oren Reshef; Luofeng Zhou
    Abstract: Generative Artificial Intelligence (AI) holds the potential to either complement knowledge workers by increasing their productivity or substitute them entirely. We examine the short-term effects of the recent release of the large language model (LLM), ChatGPT, on the employment outcomes of freelancers on a large online platform. We find that freelancers in highly affected occupations suffer from the introduction of generative AI, experiencing reductions in both employment and earnings. We find similar effects studying the release of other image-based, generative AI models. Exploring the heterogeneity by freelancers’ employment history, we do not find evidence that high-quality service, measured by their past performance and employment, moderates the adverse effects on employment. In fact, we find suggestive evidence that top freelancers are disproportionately affected by AI. These results suggest that in the short term generative AI reduces overall demand for knowledge workers of all types, and may have the potential to narrow gaps among workers.
    Keywords: generative AI, large language model (LLM), online labor market
    Date: 2023
  14. By: Yael Hochberg; Ali Kakhbod; Peiyao Li; Kunal Sachdeva
    Abstract: Implementing a state-of-the-art machine learning technique for causal identification from text data (C-TEXT), we document that patents authored by female inventors are under-cited relative to those authored by males. Relative to what the same patent would be predicted to receive had the lead inventor instead been male, patents with a female lead inventor receive 10% fewer citations. Patents with male lead inventors tend to undercite past patents with female lead inventors, while patent examiners of both genders appear to be more even-handed in the citations they add to patent applications. For female inventors, market-based measures of patent value load significantly on the citation counts that would be predicted by C-TEXT, but do not load significantly on actual forward citations. The under-recognition of female-authored patents likely has implications for the allocation of talent in the economy.
    JEL: C13 J16 J24 J71 O30
    Date: 2023–08
  15. By: Nicolas Ameye; Jacques Bughin; Nicolas van Zeebroeck
    Date: 2023–08–25
  16. By: Ilias Filippou; James Mitchell; My T. Nguyen
    Abstract: Using close to 40 years of textual data from FOMC transcripts and the Federal Reserve staff's Greenbook/Tealbook, we extend Romer and Romer (2008) to test if the FOMC adds information relative to its staff forecasts not via its own quantitative forecasts but via its words. We use methods from natural language processing to extract from both types of document text-based forecasts that capture attentiveness to and sentiment about the macroeconomy. We test whether these text-based forecasts provide value-added in explaining the distribution of outcomes for GDP growth, the unemployment rate, and inflation. We find that FOMC tales about macroeconomic risks do add value in the tails, especially for GDP growth and the unemployment rate. For inflation, we find value-added in both FOMC point forecasts and narrative, once we extract from the text a broader set of measures of macroeconomic sentiment and risk attentiveness.
    Keywords: monetary policy; sentiment; uncertainty; risk; forecast evaluation; FOMC meetings; textual analysis; machine learning; quantile regression
    JEL: F31 G11 G15
    Date: 2023–08–30
  17. By: Fabian Heymann (SFOE - Swiss Federal Office of Energy); Konstantinos Parginos (PERSEE - Centre Procédés, Énergies Renouvelables, Systèmes Énergétiques - Mines Paris - PSL (École nationale supérieure des mines de Paris) - PSL - Université Paris sciences et lettres - CNRS - Centre National de la Recherche Scientifique); Ali Hariri (EPFL - Ecole Polytechnique Fédérale de Lausanne); Gabriele Franco (PANETTA Law Firm)
    Abstract: The growing prevalence and potential impact of artificial intelligence (AI) on society rises the need for regulation. In return, the shape of regulations will affect the application potential of AI across all economic sectors. This study compares the approaches to regulate AI in the European Union (EU), the United States (US) and China (CN). We then apply the findings of our comparative analysis on the energy sector, assessing the effects of each regulatory approach on the operation of a AI-based short-term electricity demand forecasting application. Our findings show that operationalizing AI applications will face very different challenges across geographies, with important implications for policy making and business development.
    Keywords: Artificial Intelligence, energy policy, load fore- casting, regulation
    Date: 2023–10–23
  18. By: Mathias, Lea
    Abstract: Multi-level Fusion in Deep Convolutional Networks for Enhanced Image Analysis
    Date: 2023–08–10
  19. By: Abdulrezzak Zekiye; Semih Utku; Fadi Amroush; Oznur Ozkasap
    Abstract: Cryptocurrencies have become a popular and widely researched topic of interest in recent years for investors and scholars. In order to make informed investment decisions, it is essential to comprehend the factors that impact cryptocurrency prices and to identify risky cryptocurrencies. This paper focuses on analyzing historical data and using artificial intelligence algorithms on on-chain parameters to identify the factors affecting a cryptocurrency's price and to find risky cryptocurrencies. We conducted an analysis of historical cryptocurrencies' on-chain data and measured the correlation between the price and other parameters. In addition, we used clustering and classification in order to get a better understanding of a cryptocurrency and classify it as risky or not. The analysis revealed that a significant proportion of cryptocurrencies (39%) disappeared from the market, while only a small fraction (10%) survived for more than 1000 days. Our analysis revealed a significant negative correlation between cryptocurrency price and maximum and total supply, as well as a weak positive correlation between price and 24-hour trading volume. Moreover, we clustered cryptocurrencies into five distinct groups using their on-chain parameters, which provides investors with a more comprehensive understanding of a cryptocurrency when compared to those clustered with it. Finally, by implementing multiple classifiers to predict whether a cryptocurrency is risky or not, we obtained the best f1-score of 76% using K-Nearest Neighbor.
    Date: 2023–08
  20. By: Sushant Acharya; William Chen; Marco Del Negro; Keshav Dogra; Aidan Gleich; Shlok Goyal; Donggyu Lee; Ethan Matlin; Reca Sarfati; Sikata Sengupta
    Abstract: We provide a toolkit for efficient online estimation of heterogeneous agent (HA) New Keynesian (NK) models based on Sequential Monte Carlo methods. We use this toolkit to compare the out-of-sample forecasting accuracy of a prominent HANK model, Bayer et al. (2022), to that of the representative agent (RA) NK model of Smets and Wouters (2007, SW). We find that HANK’s accuracy for real activity variables is notably inferior to that of SW. The results for consumption in particular are disappointing since the main difference between RANK and HANK is the replacement of the RA Euler equation with the aggregation of individual households’ consumption policy functions, which reflects inequality.
    Keywords: HANK model; Heterogeneous-agent New Keynesian (HANK) model; Bayesian inference; sequential Monte Carlo methods
    JEL: C11 C32 D31 E32 E37 E52
    Date: 2023–08–01
  21. By: Abdul-Lateef Haji-Ali; Jonathan Spence
    Abstract: We consider the problem of estimating a nested structure of two expectations taking the form $U_0 = E[\max\{U_1(Y), \pi(Y)\}]$, where $U_1(Y) = E[X\ |\ Y]$. Terms of this form arise in financial risk estimation and option pricing. When $U_1(Y)$ requires approximation, but exact samples of $X$ and $Y$ are available, an antithetic multilevel Monte Carlo (MLMC) approach has been well-studied in the literature. Under general conditions, the antithetic MLMC estimator obtains a root mean squared error $\varepsilon$ with order $\varepsilon^{-2}$ cost. If, additionally, $X$ and $Y$ require approximate sampling, careful balancing of the various aspects of approximation is required to avoid a significant computational burden. Under strong convergence criteria on approximations to $X$ and $Y$, randomised multilevel Monte Carlo techniques can be used to construct unbiased Monte Carlo estimates of $U_1$, which can be paired with an antithetic MLMC estimate of $U_0$ to recover order $\varepsilon^{-2}$ computational cost. In this work, we instead consider biased multilevel approximations of $U_1(Y)$, which require less strict assumptions on the approximate samples of $X$. Extensions to the method consider an approximate and antithetic sampling of $Y$. Analysis shows the resulting estimator has order $\varepsilon^{-2}$ asymptotic cost under the conditions required by randomised MLMC and order $\varepsilon^{-2}|\log\varepsilon|^3$ cost under more general assumptions.
    Date: 2023–08

This nep-cmp issue is ©2023 by Stan Miles. It is provided as is without any express or implied warranty. It may be freely redistributed in whole or in part for any purpose. If distributed in part, please include this notice.
General information on the NEP project can be found at For comments please write to the director of NEP, Marco Novarese at <>. Put “NEP” in the subject, otherwise your mail may be rejected.
NEP’s infrastructure is sponsored by the School of Economics and Finance of Massey University in New Zealand.