Quant LetterNo. 55
127 items across 9 sections, as sent to readers on 3 July 2024. Paper titles open their ML-Quant page; ↗ goes to the source.
Quantitative-finance and ML-for-finance preprints from arXiv.
12 items
The research investigates the concept of dependence uncertainty and its effect on tail risk measures in relation to credit risk, showing that even minor positive dependence between losses can lead to perfectly correlated tails beyond a certain point.
6 shares3 citations todaySource ↗
The article examines a loss-averse version of the multiplicative habit formation preference, establishing the optimal consumption and investment policies and providing a detailed analysis of the HJB equation for the concavified problem.
3 shares1 citation todaySource ↗
The evaluation of the M6 Competition's performance assessment methods shows that most participants generated lower risk-adjusted returns and failed to outperform in raw returns, emphasizing the differing incentives between competitors and professional investors.
2 shares1 citation todaySource ↗
The research uses double machine learning to show that credit ratings significantly influence a company's leverage ratio, with the impact varying based on the rating, and the shift from no impact to a positive impact is gradual.
8 sharesSource ↗
The research examines a paradox in single transferable vote elections, where increased participation from non-supporters or decreased participation from supporters can alter the election result, using Scottish elections as examples.
5 sharesSource ↗
The study introduces six new indices to measure competitive balance in the UEFA Champions League group stage, finding no evidence of any trend in the competitive balance from the 2003/04 to 2023/24 seasons.
4 shares10 citations todaySource ↗
The research explores the effects of the European Central Bank's Anti-Fragmentation Policy on non-ERM II, EU countries, finding significant evidence of non-linearities in the international spillovers of ECB policy.
2 sharesSource ↗
Revolutionizing Resource Allocation: The article discusses the Global Compute Exchange (GCX), a new platform that uses blockchain technology to create a secure marketplace for trading computational power. The platform aims to optimize resource use, stabilize pricing, and make access to computational resources more democratic.
5 shares3 citations todaySource ↗
The study introduces LSTM-ARIMA, an algorithmic investment strategy that combines LSTM and ARIMA models, and proves its superior performance in risk-adjusted return measures across three equity indices.
9 shares61 citations todaySource ↗
The article presents an alternative proof showing that a finite continuous non-decreasing submodular set function can be expressed as a supremum of measures dominated by the function, using a standard extension theorem of measures.
6 sharesSource ↗
Formulaic Alpha Generation Framework: The paper introduces AlphaForge, a two-stage formulaic alpha generating framework that uses a generative-predictive neural network for factor generation and dynamic weight adjustment, showing improved performance in alpha factor mining.
6 shares40 citations todaySource ↗
The paper expands Merton's structural model for public companies, assuming observed liabilities, and provides formulas for risk-neutral equity and liability values, default probabilities, and machine learning estimators of the model's parameters.
5 sharesSource ↗
Working papers in finance and economics from SSRN.
18 items
A study shows that a ban on margin trading in cryptocurrencies leads to less market volume and volatility, but more informed trading on Binance.
3 sharesSource ↗
A new concept called approximate mixing for random variables on metric spaces provides a balance between traditional mixing assumptions and proves a central limit theorem for nonstationary time series on Hilbert spaces.
9 sharesSource ↗
Hedge funds perform worse when there is economic policy uncertainty, especially those with high EPU beta funds due to their inability to predict EPU fluctuations.
5 sharesSource ↗
A comparison study reveals that option-implied volatility underestimates risk during market turbulence when estimating risk factor’s volatility in Value-at-Risk (VaR) computation.
2 sharesSource ↗
The PLOD algorithm, based on the BOD method, can accurately predict the desired utility function from data without needing the exact utility function, enhancing efficiency in data science and analytics.
2 sharesSource ↗
The article questions the validity of the MeanVariance (MV) model and the Capital Asset Pricing Model (CAPM) due to market portfolio inefficiency and the failure of asset expected returns in beta under realistic constraints.
2 sharesSource ↗
The article extends the Gibson and Schwartz 1990 and Schwartz and Smith 2000 twofactor models for commodity spot price to include stochastic volatility and correlation, improving match with volatility smiles and studying timevarying correlation in commodity markets.
2 sharesSource ↗
The article proposes a data-driven framework using a hybrid model of Support Vector Machine and Deep Neural Network for predicting the ballistic performance of composite armor, demonstrating high accuracy and generalizability.
2 sharesSource ↗
The article introduces a multiscale neural network framework that incorporates selfsimilarity as prior knowledge, enabling the modeling of selfsimilar dynamical systems in complex systems.
3 sharesSource ↗
The article proposes an AI-based multimodal auditing system that integrates data from diverse documents and allows auditors to complete tasks using natural language, demonstrating its feasibility through a simulation case study.
2 sharesSource ↗
The increased trading of SP 500 index options that expire on the same day reduces stock market volatility due to market makers' intraday rebalancing of the index.
6 sharesSource ↗
The research provides an in-depth review of deep learning techniques for text summarization, including different structures, attention mechanisms, and recent evaluation metrics.
2 sharesSource ↗
Two machine learning models, MLPRegressor and Keras Sequential, have been created to predict the synthetic accessibility scores of molecules with a mean squared error of around 0.20.
2 sharesSource ↗
The research suggests using deep neural networks to identify deception attacks in cyber-physical systems that disrupt system performance by injecting false data.
2 sharesSource ↗
The article presents a new Esscher-based algorithm for determining default probabilities in structural credit risk models, providing a more accurate tail behavior than the traditional Merton model.
2 sharesSource ↗
The study examines the possible profits and risks of basis trading with Ethereum futures contracts after the launch of Ethereum spot ETFs, and suggests alternative trading strategies and further research on long-term profitability.
2 sharesSource ↗
The research explores investor knowledge of startup IPOs in Karnataka, using statistical analysis to understand the link between IPO size, company profile, sector performance, and investment choices.
3 sharesSource ↗
The study finds that bond peer momentums, particularly those with shared analyst linkage, can predict future bond returns in the corporate bond market. This strategy can yield significant profits, especially for bonds with less investor attention and higher arbitrage costs.
3 sharesSource ↗
Economics working papers from RePEc's NEP field reports.
25 items
The research shows machine learning models effectively predict stock price movements during unstable periods like the Russia-Ukraine war, but not during stable periods.
33 sharesSource ↗
The paper finds no significant arbitrage opportunities in copper futures traded on the Shanghai Futures Exchange from 2006 to 2022, supporting the market efficiency rule.
31 sharesSource ↗
The study reveals that different models are more effective for different assets like gold, cocoa, and the S&P500 index, impacting risk management strategies.
31 sharesSource ↗
The paper finds that positive market returns at the high-frequency level increase price volatility in the cryptocurrency market, contradicting traditional financial literature.
25 sharesSource ↗
The research shows volatility clustering in the return series of crude oil, Bitcoin, and selected Asian stock exchanges, with varying degrees of short and long-term volatility spillover.
24 sharesSource ↗
The article introduces a new, faster test for alpha in linear factor pricing models, which remains valid even with a large number of securities and can account for some pricing errors, finding evidence against certain asset pricing models during the Great Recession.
23 sharesSource ↗
The study reveals that hedge funds significantly benefit from the low beta anomaly, attributing about 2.3% per year of apparent alpha to the anomaly rather than manager skill, with low skill managers relying most on the anomaly for returns.
22 sharesSource ↗
The study presents a new duality theory for optimal consumption-investment problem, using alternative data like social media commentary and pandemic data for predicting stock trends, and offers a unique smooth solution for an agent with constant relative risk aversion.
22 sharesSource ↗
The research indicates that using pricing factor variables in the corporate valuation process with an eight-factor model reduces estimation errors and improves precision, contributing to the intersection of empirical asset pricing and corporate finance.
21 sharesSource ↗
The paper examines the roughness of oil market volatility using unspanned stochastic volatility models, demonstrating that an additional parameter indicating the roughness improves the calibration by almost a factor of 10, highlighting the importance of modeling the time dependency of the roughness.
19 sharesSource ↗
The study uses machine learning to enhance the accuracy of Loss Given Default (LGD) estimation in situations with limited cash-flow data, using a European mortgage portfolio.
25 sharesSource ↗
The research finds that despite higher risks, machine learning algorithms can select a subset of stocks that outperform the S&P 500, with the importance of determining factors changing over time.
24 sharesSource ↗
The paper finds that machine learning models are more effective than regression models at modeling credit spread changes, and applies this to US and Euro Area bonds to measure the impact of different economic and financial variables.
19 sharesSource ↗
Machine learning models, particularly regression-tree methods, are better at pricing cryptocurrency options than traditional models due to their adaptability to cryptocurrency market dynamics.
45 sharesSource ↗
Interpretable machine learning proves superior in modeling corporate bond recovery rates, offering valuable insights into recovery rate determinants.
39 sharesSource ↗
Machine learning models, specifically random forests and bagging, outperform traditional models in predicting the US stock market direction using volatility indices.
26 sharesSource ↗
Machine learning models effectively predict cryptocurrency market returns using factors like market price, past alpha illiquidity, and momentum, and remain profitable despite high portfolio turnover.
25 sharesSource ↗
The Random Forest machine learning algorithm successfully predicts dividend payouts in the Vietnamese stock market, enhancing decision-making and market transparency.
23 sharesSource ↗
Ridge, Elastic Net, and SVR machine learning methods outperform others in nowcasting GDP, proving more effective than principal components regression.
21 sharesSource ↗
The study uses an advanced deep learning model to predict Bitcoin's volatility, enhancing prediction accuracy by nearly 35% compared to other models. It also offers a risk warning system and trading strategy for investors.
21 sharesSource ↗
The research uses machine learning to determine the predictive power of liquidity variables in economic fluctuations, with private sector data proving more useful than public sector data.
34 sharesSource ↗
The study examines the volatility puzzle in China's stock market, concluding that individual investor attention increases volatility and securities analysts can lessen market information asymmetry.
16 sharesSource ↗
The research uses machine learning to measure the gap between research and practice in accounting academia, identifying a significant topic-related gap that is more evident in the USA than in Europe.
6 sharesSource ↗
The article explores the likelihood of employers hiring older workers during labor shortages, revealing that recruitment difficulties increase preferences for hiring retired individuals and those over 55 with experience-related qualities.
4 sharesSource ↗
The paper was incomplete and did not provide sufficient information for a summary.
2 sharesSource ↗
The general machine-learning papers the letter carried in 2023-25.
19 items
A new method for creating synthetic data uses 1 billion diverse personas, potentially transforming large language model research and development.
258 shares466 citations todaySource ↗
UnSAM, a new automatic image segmentation model, performs competitively without needing human annotations, surpassing previous unsupervised models.
97 shares32 citations todaySource ↗
PoliFormer, an indoor navigation agent trained in simulation, performs well in real-world scenarios, showing potential for various applications.
61 shares85 citations todaySource ↗
A proposed agent engineering framework offers a scalable and flexible workflow for multiple autonomous agents to collaborate across various domains.
13 shares4 citations todaySource ↗
A new task called dataset size recovery aims to determine the number of samples used to train a model, with a method called DSiRe proposed for this purpose.
12 shares9 citations todaySource ↗
The study finds Small Multilingual Language Models (SMLM) excel in zero-shot cross-lingual sentiment analysis, while Large Language Models (LLM) perform better in few-shot scenarios.
11 shares15 citations todaySource ↗
3D Scene Gen: HouseCrafter, a new method, can transform a floorplan into a 3D indoor scene using a 2D diffusion model trained on large-scale web images.
7 shares10 citations todaySource ↗
The paper suggests using the Gittins index from the Pandora's Box problem in economics as a function for cost-aware Bayesian optimization.
7 shares22 citations todaySource ↗
The study reveals that Large Language Models with more layers are more robust, and identifies four universal stages of inference across different models.
7 shares153 citations todaySource ↗
AutoCap and GenAu, two new models for generating ambient sounds and effects, are introduced, improving the quality of audio captions and generated audio.
6 shares40 citations todaySource ↗
File Byte Transformers: ByteFormer is a deep learning model that enhances image classification accuracy by 5% and can perform audio classification without specific preprocessing, showcasing its versatility.
156 shares19 citations todaySource ↗
InstructGPT, a large language model, can reduce diversity in collaborative writing, making different authors' writings more similar and decreasing overall content variety.
149 shares244 citations todaySource ↗
Large Language Models (LLMs) inaccurately perceive human decision-making as more rational than it is, aligning more with expected value theory, despite seeming to mimic human behavior.
133 shares63 citations todaySource ↗
Concept forgetting in AI models can be significantly reduced by a new fine-tuning method called LDIFS, which helps retain pre-trained knowledge while working on different tasks.
75 shares102 citations todaySource ↗
Language Model Optimization: Automix is a new method that saves computational cost by over 50% by directing queries to larger language models based on the accuracy of outputs from smaller models.
41 shares129 citations todaySource ↗
The study shows that anonymized external data can mitigate the loss of diagnostic accuracy in speech-based COVID-19 diagnosis systems caused by voice anonymization methods.
26 shares7 citations todaySource ↗
Buggy Defenses: The paper exposes major flaws in the Sabre defense against adversarial examples, including gradient masking and bugs in the evaluation code, which when corrected, drop Sabre's robust accuracy to 0%.
24 shares3 citations todaySource ↗
The study successfully demonstrates the use of embedded field programmable gate array (eFPGA) technology in machine learning for data pipelines in future collider experiments, achieving perfect accuracy.
23 shares9 citations todaySource ↗
The research investigates the explainability of machine learning in forensic authorship profiling, finding that lexical features and place names play a significant role in the classification process.
20 shares4 citations todaySource ↗
Papers that shipped their code, from the Papers with Code feed (2023-25).
7 items
The first article highlights the significance of enhancing multidimensional transformers for handling long sequences in different fields.
1,240 shares
The second article introduces a technique that can successfully boost the loglikelihood of a selected response while reducing that of a discarded one, a challenge not easily overcome by symmetric pairwise loss.
177 shares
The third article shows how Grendel enhances rendering quality by increasing 3DGS parameters across several GPUs in large-scale, high-resolution scenes.
124 shares
Longchain Reasoning for LLMs: Large Language Models (LLMs) have difficulty with mathematical reasoning due to the requirement for detailed and accurate reasoning chains.
62 shares
Fewer Learning Rates, Better Results: By dividing parameters into blocks and assigning a single optimal learning rate to each, over 90% of learning rates could be eliminated without causing damage.
61 shares
MultiGranularity Visual Tuning: A high-resolution visual encoder is suggested to capture detailed visuals, which are then combined with basic visual features through a ConvGate fusion network.
50 shares
The process of increasing the size and complexity of deep Reinforcement Learning (RL) methods poses a significant challenge.
37 shares
Repositories the letter featured.
10 items
The article offers a detailed guide on using machine learning methods for analyzing time series data.
883 shares
The piece explores a trading system that utilizes algorithms for its operations.
655 shares
The article explains the process of setting up a basic Deep Hedging engine.
169 shares
The piece discusses how to use exchange calendars with pandas for trading applications.
747 shares
The article introduces a training structure for Stable Baselines3 reinforcement learning agents, featuring hyperparameter optimization and pretrained agents.
1,884 shares
The article provides a collection of guides and examples for utilizing Google's Gemma open models.
99 shares
The article explores a PyTorch version of the Transformer model, as detailed in the Attention is All You Need paper.
8,578 shares
The article presents a robust and scalable Python library designed specifically for detecting anomalies and outliers.
8,230 shares
The piece introduces an autoregressive character-level language model designed to boost productivity.
2,259 shares
Industry news: funds, hiring, markets and regulation.
20 items
Jonathan Xiong, former co-CEO of Millennium’s Asia operations, has started a new hedge fund, Arrowpoint Investment Partners, with over $1bn in assets.
5 shares
Falcon Investment Management in London has introduced a new fund, offering select investors temporary access to its exclusive investment platform.
4 shares
Goldman Sachs data reveals that hedge funds are shifting away from technology stocks, particularly semiconductor and chip equipment, after strong gains in Q1 2024.
3 shares
The insider trading case involving Segantii Capital Management, Simon Sadler, and ex-trader Daniel La Rocca has been delayed to 15 October by a Hong Kong court.
3 shares
Digital asset investment products saw a third week of outflows, but the total of $30m is a significant decrease in recent outflows, as per CoinShares' report.
3 shares
Mind Money, a European broker, offers various trust management methods and represents major asset management firms in Europe, says CEO Julia Khandoshko.
3 shares
Taproot Management, a new hedge fund, plans to replace costly portfolio managers with a team of analysts for its $500m launch next year, Bloomberg reports.
3 shares
The European Energy Exchange and European Commodity Clearing are offering a six-month fee waiver for trading EEX Nordic Zonal Power Futures and EEX Nordic System Price Power Futures.
3 shares
BlackRock plans to acquire UK-based private markets data provider Preqin for £2.55bn ($3.2bn), merging Preqin’s data and research tools with Aladdin’s workflow capabilities.
3 shares
Hedge funds sold global equities at the fastest rate in two years in June, marking the third month of net selling, mainly driven by short sales, says a Reuters report.
2 shares
In 2024, new hedge fund launches surged and liquidations remained steady, pushing industry assets to a record $4.3tn due to increased geopolitical risk, according to the HFR Market Microstructure Report.
2 shares
Odey Asset Management, owned by Crispin Odey, reported a £37.1m goodwill impairment following its closure, as per a Bloomberg report.
2 shares
Hedgeweek's European Digital Assets Summit is a one-day event focusing on the development of digital asset strategies for European hedge fund operational and product specialists.
2 shares
Pershing Square, a hedge fund led by Bill Ackman, plans to sell shares at $50 each in its upcoming New York IPO, with the listing date still unconfirmed, as per a BayStreet report.
2 shares
Baupost Group, led by Seth Klarman, has reduced about 19% of its investing staff in a major restructuring, marking the largest staff cut in the hedge fund's 42-year history, according to Bloomberg.
2 shares
Abu Dhabi's Al Maryah Island is facing an office space shortage due to the successful attraction of top hedge funds.
2 shares
The IRS has apologized to Citadel founder Ken Griffin for the unauthorized release of his tax returns.
1 shares
Citadel and Millennium, two hedge fund giants, have reported strong gains in the first half of the year.
1 shares
Job opportunities are being created within Citi due to its disorganized data.
1 shares
Hedgeweek has announced the winners of the Global Digital Assets Awards 2024, recognizing excellence in digital assets.
0 shares
Episodes on markets, quant methods and economics.
10 items
Financial journalist Herb Greenberg talks about market evolution, his investigations into financial frauds like Herbalife, and his new online investment media platform, Wall Street Beats.
10 shares
In a podcast, Arindam Sandilya, Lorenzo Ravagli, Ladislav Jankovic, and Juan Duran-Vara discuss the H2 outlook for FX Derivatives.
9 shares
Dave Mabe, CTO at Trade Ideas and a successful independent trader, is featured on the Confessions of a Market Maker podcast.
6 shares
Strategists Francis Diamond, Raphael Brun-Aguerre, and Aditya Chordia discuss the implications of the first round of voting in the French elections on French spreads and European fixed income markets.
6 shares
Teresa Ho and Pankaj Vohra, Short Duration strategists, discuss the growth of the stablecoin market and its potential impact on the Treasury market.
5 shares
In a podcast recorded on 28 June 2024, Meera Chandan, Arindam Sandilya, Junya Tanase, Antonin Delair, James Nelligan, and Patrick Locke discuss the key macro FX themes for the second half of the year.
5 shares
Ben Bennett discusses navigating election uncertainty, concentration risk, and the effects of growing deficits on global productivity in a podcast hosted by Frances Watson.
4 shares
Michael Sonenscher's book Capitalism: The Story Behind the Word delves into the history and evolution of capitalism, with a focus on public debt, war finance, and labor division.
4 shares
Erik Townsend and Patrick Ceresna of MacroVoices host WindShift Capital Founder Bill Blain to discuss the equity market, inflation, and bond yields.
4 shares
Kathleen McCarragher discusses her investment experiences, views on AI and growth opportunities, and advice on team building and leadership in a conversation with Goldman Sachs’ Betsy Gorton.
4 shares
Posts from quant and economics blogs and newsletters.
6 items
The Schaff Trend is a trading tool that offers day traders superior tracking of market fluctuations and price trends.
2 shares
Prediction market traders have been contemplating the chance of a different candidate winning the November election, aside from Joe Biden or Donald Trump.
0 shares
The author, a Physics and Applied Maths major at UCT, shares their increasing realization of opportunities for further study in England and America during their third year.
0 shares
The author shares their experience studying Physics and Applied Maths at UCT, which sparked their interest in pursuing further studies abroad.
0 shares
The first presidential debate has reportedly shaped public opinion about the potential results of the upcoming November election and the Democratic National Convention.
0 shares
The author stresses the uniqueness of each market and urges readers to conduct their own research.
0 shares