Research Directions
Backtesting frameworks, out-of-sample testing, and common biases: why backtests look good but fail in live trading.
48 related activitiesLast 30 days 31 itemsFeatured among them 1 items
This paper proposes PALM (Point-in-time Adaptation for financial Language Models) as an alternative to annual pre-training of financial language models. It achieves adaptation by fitting low-rank adapters on permitted financial text without modifying any pre-trained weights.
Read originalThis arXiv paper defines the recovery of tick-level queue history from aggregated L2 limit order book snapshots as a conditional partial identification problem. It constructs a path-preserving compiler that only varies among three cancellation allocation rules—front-of-queue, quantity-weighted random, and back-of-queue—while fixing the aggregated path, reconciled market removals, latent order partitioning, and new order arrivals.
Read originalThis arXiv paper evaluates the reliability and investment consequences of TimeGAN-generated scenarios in Rachev ratio portfolio allocation. Using 12 ETFs, it compares historical, substitution, and hybrid strategies under delayed execution and transaction costs. The study finds that generated scenarios underestimate volatility and tail losses and distort dependencies between assets; extending training only reduces reconstruction loss without improving cross-asset correlation accuracy. Substitution strategies increase turnover and are sensitive to the lower bound of the objective function denominator, and adjusting this lower bound does not consistently improve realized downside risk.
Read originalQuantConnect Lean commit number 18197 adds the ShareClassMarketCap field, with MarketCap still retained as the company's total market capitalization. This commit is verified by GitHub signature and co-authored by Claude Fable 5.1.
Read originalvectorbt v1.2.0 has been officially released. This version introduces the ulcer index and Martin ratio indicators, adds the nbformat>=4.2.0 dependency, and fixes multiple issues including OBV calculation, EWM standard deviation ddof handling, portfolio P&L expectation statistics, and drawdown logging.
Read originalA former equity strategy partner and independent investment firm operator used AI in retirement to backtest his long-standing personal strategy, which combined Peter Lynch's growth stock selection, Hilary Kramer's undervalued obscure stock selection, and earnings revision discipline. He corrected for survivorship bias, used point-in-time data, calculated pre-fee returns, and compared it against random stock selection from the same investment pool. The results showed that the strategy did not outperform random selection, with ordinary index funds performing better than the strategy.
Read originalQuantocracy has compiled the latest quant research-related links as of October 18, 2026, covering topics including trend following, futures continuous contract construction, in-sample to out-of-sample Sharpe ratio decay, the impact of liquidity on stock returns and option pricing, and cryptocurrency cross-sectional trend factors. One study based on 12 commodity futures found that the expected retention ratio of in-sample Sharpe ratios from a 10-year backtest on new data is 30%.
Read originalThis arXiv paper proposes a NoLBERT encoder pre-trained only on texts from 1976 to 1995, comparing it with LDA, MiniLM, and FinBERT on 1.26 million news articles from 2012 to 2024 to predict returns of the 1000 largest U.S. stocks.
Read originalZipline Reloaded 3.2.0 is a maintenance release. It supports Python 3.11–3.14, removes support for Python 3.10, and is not yet compatible with Pandas 3 or Python 3.15. The pandas version remains constrained to below 3.0.
Read originalCode change commit number 18193 in QuantConnect Lean migrates BaseExtendedDictionary, ReadOnlyExtendedDictionary.
Read originalQuantConnect Lean Release 18192 migrates BaseExtendedDictionary, ReadOnlyExtendedDictionary, and SessionConsolidator.
Read originalQuantConnect Lean released commit 18189 (corresponding hash 1a91415), which has a verified signature on GitHub. This update hides FundamentalProperty and the Get methods that accept this parameter from the code stubs.
Read originalQuantocracy compiled recent quant research links on October 15, 2026. This roundup includes Allocate Smartly's backtest of the OptimumV strategy from Financial Mentor (backtest period starting in 1987).
Read originalNautilusTrader 2.0.0rc7 was released on October 15 (UTC), removing Python 3.12 support and adding compatibility with Python 3.13 and 3.14. This update adds features such as negative price support for futures, local calendar-based scheduled dispatch, batch order book incremental replay, and CME Globex option candidate IDs, while also including multiple breaking changes to interfaces and configurations.
Read originalQuantocracy released its latest roundup of quantitative research links on October 14, 2026, featuring multiple research pieces and content related to the quantitative field. These include an out-of-sample evaluation study of Hidden Markov Model (HMM) market state identification, an analysis of the impact of complexity on strategy performance, and research on cross-sectional momentum ranking methods. The related research covers backtest results across different sample periods and discussions of strategy logic.
Read originalQuantpedia released a study comparing four four-state Gaussian HMMs that only differ in training start year (1970, 1980, 1990, 2000), evaluating market state recognition performance on the same 2010-2023 test set.
Read originalAuthor Kris Longmore shares content from his 72-minute interview on The Algorithmic Advantage podcast, looking back on an early experience at a Sydney traders' meetup where he was unable to answer a proprietary trader's question about his own trading edge. He later joined that proprietary firm as a researcher and partner, and summarized relevant insights after leaving to trade independently.
Read originalQuantocracy compiled recent quantitative content on October 12, 2026, covering an introduction to Quant Observer, a quantitative tools directory platform; a historical effectiveness analysis of excess returns from insider buying; a backtest of volatility clustering effects in S&P 500 constituents from 2006 to 2025; a guide to building a multi-asset, multi-timeframe trading pipeline; and AI answers on the replicability of empirical finance research.
Read originalAlpha Architect interprets Dmitriy Muravyev's 2026 working paper, which uses an in-house LLM replication pipeline to reconstruct the core results of 1,328 papers published in the Journal of Finance, Journal of Financial Economics, and Review of Financial Studies between 2000 and 2020, of which 1,005 were eligible for subsequent out-of-sample testing.
Read originalvectorbt v1.1.2 has been released. This update fixes multiple functional defects, including the isclose helper function returning True for any finite number, NaN consistency issues in is_sorted_nb, portfolio NaN trade plotting marker handling, ignored NaN returns in cond_value_at_risk, and intraday time retention during NumPy datetime64 conversion.
Read original
