Quant LetterNo. 32
67 items across 5 sections, as sent to readers on 9 January 2024. Paper titles open their ML-Quant page; ↗ goes to the source.
Quantitative-finance and ML-for-finance preprints from arXiv.
10 items
The study uses text mining and natural language processing to examine quantitative finance papers from 1997 to 2022 on the arXiv preprint server. It identifies topic trends, most cited researchers and journals, and compares different topic modeling algorithms.
5 shares3 citations todaySource ↗
The article introduces a new linearization of the Markowitz portfolio optimization model that reduces maximum portfolio drawdown, particularly beneficial during financial crises, and offers a quicker, more profitable version of this model.
6 sharesSource ↗
The article expands on previous work on determining feasible exchange ratios for merging companies in a volatile environment, setting both maximum and minimum limits for acceptable exchange ratios and employing a diagrammatic method for improved visualization.
5 sharesSource ↗
The article introduces a model for price dynamics in Automated Market Makers, suggesting a reference market price and deriving several analytical results about its behavior using local times and excursion-theoretic methods.
5 shares5 citations todaySource ↗
The note reexamines the SWIFT method for pricing European options under models with a known characteristic function in 2023, discussing potential enhancements and pointing out some limitations of the method.
4 shares1 citation todaySource ↗
The proportion of non-core to core funding in advanced economies' banking systems is influenced by global factors, with exchange rate flexibility providing some protection, except during significant global financial crises.
20 sharesSource ↗
The article discusses portfolio generating functions and rational option pricing, showing that a portfolio's value can replicate a function's value if the function satisfies a certain equation.
13 sharesSource ↗
The research investigates a financial market model, proving it's free of arbitrage only if the interest rate is zero, and assesses the hedging error from misrepresenting price stickiness.
13 shares1 citation todaySource ↗
The paper uses the Hamiltonian approach from quantum theory to option pricing with fluctuating interest rates, deriving pricing kernels and option prices under a specific model.
10 shares1 citation todaySource ↗
The study introduces a new formula for pricing spread call options under log-normal models, which outperforms the formula presented in a previous study for certain model parameters.
9 sharesSource ↗
Working papers in finance and economics from SSRN.
24 items
The Combinatorial Purged Cross-Validation (CPCV) method is superior in financial analytics for reducing overfitting risks, outperforming traditional methods like K-Fold and Walk-Forward.
3 shares18 citations todaySource ↗
Research using SPX options data from 2011 to 2022 found that Volterra Bergomi models do not accurately capture implied volatility due to the roughness component's structural limitations.
89 sharesSource ↗
The paper introduces a new learning algorithm that uses Topological Data Analysis to prevent local models from forgetting previous knowledge and improve server feature capture.
2 sharesSource ↗
Incorporating Risk premia strategies in multi-asset portfolios can lessen left-tail exposure, but diversification within options needs maximizing the number of volatility parameters for a direct trading strategy.
3 sharesSource ↗
Semantic fingerprinting of news headlines can measure the impact of news on major currency indices, showing a positive correlation between news intensity and currency return volatility.
2 sharesSource ↗
The article discusses a formula for managing portfolio risk using Gaussian copulas, which can track new risk metrics like implied conditional portfolio volatilities and weights.
13 sharesSource ↗
The paper assesses the use of the Shapley value in credit scoring to enhance transparency and comprehension of machine learning algorithm decisions.
2 sharesSource ↗
The study investigates the strategic behavior of algorithmic trading firms to understand their competitive advantage and market survival.
2 sharesSource ↗
The research reveals a positive link between CEOs' option-based compensation and discussions about political risk in earnings conference calls, indicating a risk-taking strategy.
3 sharesSource ↗
The study employs a hybrid machine learning method, CNN-LSTM, to predict the interconnectedness of economic sectors in emerging markets, showing its effectiveness in improving prediction accuracy.
2 sharesSource ↗
A paper suggests that exploiting predictability in style demand can yield annualized returns of 12% from both reversals and momentum, according to an examination of return autocorrelations.
3 sharesSource ↗
A modification to the base learner in bootstrap aggregation and boosting can significantly improve predictive accuracy in volatility forecasting, addressing substantial errors from parameter estimation.
2 shares2 citations todaySource ↗
Shorter rebalancing periods are more effective in capturing academic momentum in portfolios, a study on portfolio sizes, weighting schemes, and rebalancing intervals reveals.
5 shares2 citations todaySource ↗
The research paper investigates the impact of global economic events on financial markets, highlighting the importance of understanding these effects for investors, financial institutions, and policymakers.
3 sharesSource ↗
Firms with larger size, lower growth, and higher profitability are more likely to improve their ESG scores, but the portfolio doesn't generate alpha after considering its exposure to profitability and growth factors.
3 sharesSource ↗
A new tool has been created for liquidity stress testing and planning in portfolios with private assets, aiding in risk assessment and cash flow management.
2 sharesSource ↗
The Arbitrage-Free Nelson-Siegel model has been expanded to a generalized model, revealing new components in the yield-adjustment term and differences across models.
5 sharesSource ↗
Research on the Russell 3000 Index from 2008-2022 shows a positive link between stock trading volume and volatility, indicating US stocks' resilience during volatile periods.
3 sharesSource ↗
Machine learning outperforms human experts in evaluating firm quality, with the residual income valuation method proving superior to the DuPont method and extensive data mining.
2 sharesSource ↗
The beta-alpha anomaly only occurs during times of low economic uncertainty, with retail investors and active mutual funds more prone to pursue high-beta securities.
2 sharesSource ↗
Utilizing the geometric mean in asset allocation can yield higher total returns than the arithmetic mean, especially with safe haven and insurance-like assets.
3 sharesSource ↗
Private equity fund performance doesn't correlate with quantitative data like past performance, but machine learning can predict future performance using qualitative data.
2 sharesSource ↗
Dichotomy & Drivers: Research into the volatility dynamics of eight major cryptocurrencies shows distinct dynamics during market booms and downturns, linked to specific events in the crypto market and macroeconomic history.
2 sharesSource ↗
A paper suggests that simple payoff-based reinforcement learning can help achieve rational expectations equilibrium in limit order markets, with speculators mainly providing liquidity.
2 shares2 citations todaySource ↗
Economics working papers from RePEc's NEP field reports.
17 items
A new hybrid model called WT-ARIMA-LSTM has been introduced for share price index futures forecasting, offering superior accuracy and robust performance in various market conditions.
17 sharesSource ↗
The Shanghai 50 ETF index options market operates efficiently when call and put options are at-the-money, but not when the call is in-the-money and the put is out-of-the-money.
22 sharesSource ↗
The effectiveness of new factor models is determined by the construction of their investment factor, with factors based on inventory growth and accounts receivable holding most of the pricing information.
16 sharesSource ↗
A partially specified stochastic volatility model, calibrated using the dynamic programming principle and the Heston model, can predict future trends for synthetic and S&P500 data.
15 sharesSource ↗
The article discusses recent advancements in estimating large, time-varying dynamic covariance matrices, with a focus on GARCH model extensions and identifying structural breaks in large covariance structures.
12 sharesSource ↗
The paper finds no significant difference in investment outcomes when using the Hou-Xue-Zhang four-factor model versus the Fama-French five-factor model.
12 sharesSource ↗
Machine learning is used to identify politically connected firms in Czechia with 85% accuracy, suggesting its use in detecting conflicts of interest.
14 sharesSource ↗
The article discusses the evolution of general-to-specific modelling from manual to automated machine learning, addressing criticisms and its ability to handle non-stationary data.
8 sharesSource ↗
Intermediate updating scenarios in univariate time series forecasting models can achieve similar or better accuracy with less computational cost, challenging the need for constant model updates.
8 sharesSource ↗
The study shows that Affine GARCH models are more efficient in portfolio optimization compared to homoscedastic variants.
11 sharesSource ↗
The research shows that in arbitrage-free markets, there are wealth-preserving strategies that perform better than just buying and holding a risky asset.
37 sharesSource ↗
The research uses machine learning to create early warning models for financial crises, with credit growth and yield curve slope being key predictors.
25 sharesSource ↗
Peter Carr Gedenkschrift: The book honors Peter Carr's contributions to Quantitative Finance, featuring new research and personal tributes from his loved ones.
26 sharesSource ↗
The study outlines optimal trading strategies for partially informed traders under certain price dynamics, showing these strategies rely on current price and a specific type of moving average price.
24 sharesSource ↗
The article introduces a model that combines bilateral and centrally cleared trading for stress testing financial networks or optimizing portfolio transfers of a defaulted clearing member.
23 sharesSource ↗
The study uses machine learning algorithms to predict profitability direction in Europe, finding that simpler algorithms can outperform more complex ones with the right data preprocessing.
20 sharesSource ↗
The paper offers a formal definition of backtestability for a statistical function of a distribution, comparing model validation and selection methods, and introduces the concept of ridge backtests.
19 sharesSource ↗
The general machine-learning papers the letter carried in 2023-25.
6 items
A novel method for creating synthetic functional training data for deep learning solutions to partial differential equations (PDEs) is proposed, eliminating the need for a numerical PDE solver and potentially broadening the scope for developing neural PDE solvers.
17 shares15 citations todaySource ↗
The study suggests a method to modify basic vision-language models like CLIP for specific tasks using text data from large language models, allowing for easy application to new classes and datasets.
95 shares63 citations todaySource ↗
Small Open-Source Language Model: The article presents TinyLlama, a compact 1.1B language model that performs remarkably well in various tasks despite its small size, having been pretrained on around 1 trillion tokens.
82 shares902 citations todaySource ↗
A Review: The article reviews machine learning models for creating synthetic data, discussing their uses, methods, privacy issues, fairness, and future research opportunities in fields like computer vision, speech, natural language processing, healthcare, and business.
38 shares309 citations todaySource ↗
The study examines the control properties of Kernel Stein discrepancies (KSDs) in distributional approximation, and presents conditions for alternative diffusion KSDs to control convergence, contributing to the first KSDs that characterize q-Wasserstein convergence.
47 shares15 citations todaySource ↗
The process of learning neural networks through gradient descent is complex, despite the advantages of integrating known symmetries, as indicated by lower bounds for various network types.
18 shares18 citations todaySource ↗
Repositories the letter featured.
10 items
Portfolio Optimization Library: A new Python library has been created for portfolio optimization using scikitlearn.
96 shares
Quantitative Finance Lecture Series: A detailed lecture series for learning quantitative finance, adapted from the Quantopian Lecture Series, is now available.
149 shares
Tulip Chart Python Bindings: Python bindings for Tulip Charts can be found in the unmaintained Tulipy Financial Technical Analysis Indicator Library.
303 shares
Causal Discovery Code for Time Series: The PCMCIΩ algorithm from the NeurIPS23 paper has been implemented for causal discovery in semi-stationary time series.
7 shares
MyShell has launched a new technology that can instantly clone voices.
6,128 shares
The article offers a detailed list of free resources for learning about machine learning.
99 shares
The article discusses a Python-based method to simulate series using the ARFIMA process.
16 shares
Xiaohongshu Works Collection Tool: The article introduces a free, open-source tool for collecting images and videos from Xiaohongshu, based on the AIOHTTP module.
2,133 shares
TinyLlama is a new compact language model, pretrained on about 1 trillion tokens for roughly 3 epochs.
4,697 shares
TRIM ELECT SIGN & MERGE TIESMerging is a new method proposed for merging models, introducing three steps to resolve conflicts and align parameters.
953 shares