AQR Tax-Shield Strategy Drives US Margin Debt to $1.45T
Today's Highlights
01 Hong Kong’s Q3 IPOs Raise HK$385.6 Billion, Surpassing Previous 15-Year Annual Totals
In the first three quarters of 2026, 115 companies went public, driving fundraising volume up 105% year-over-year. The hardware equipment sector led with HK$188.1 billion in proceeds. Deloitte forecasts approximately 160 new listings and at least HK$480 billion in fundraising for the full year, while KPMG suggests the total could approach HK$500 billion, setting a new historical record. CICC leads underwriting volume at HK$910 billion, while Futu Securities leads in the number of underwritten deals (67). 1
02 Tech Stock Buybacks Draw Attention
The Securities Times reviewed the top ten market stories of the week, highlighting the intense wave of tech stock buybacks. Multiple leading A-share tech companies have centrally announced buyback plans. Aggregating the disclosed buyback amounts reveals a substantial scale, reflecting industrial capital’s intent to stabilize the market. For details, see the cited link. 2
03 Major Update to Fund Practitioner Training Syllabus
The Fund Practitioner Follow-up Professional Training Syllabus (2026 Revision) was released two days ago, marking the most significant structural update in a decade. Key changes include:
- Tighter Credit Hours: The annual total remains at no less than 15 hours. However, the minimum hours for integrity, discipline, and law have increased from 5 to 6, and live-stream training has been officially included.
- Framework Restructuring: The old version’s 7 modules have been upgraded into three major sections—“Industry Fundamentals,” “Securities Investment Funds,” and “Equity Investment Funds”—comprising 15 parts.
- Content Expansion: New topics include the “Five Essentials and Five Don’ts” financial culture, custody, public opinion, and AI frontiers. Securities-related modules now cover public fund investment advisory, quantitative and passive investing, and performance fees. For the first time, a dedicated equity section covers venture capital, M&A, fund-of-funds, and government investment funds. 3
04 Tax-Aware Hedging Strategies Drive US Margin Loans to $1.45 Trillion
According to the Wall Street Journal, demand for tax-efficient long-short strategies offered by hedge funds like AQR and Quantinno has surged, driving a spike in US margin loans. Annualized metrics for margin businesses at Morgan Stanley, Bank of America, and Goldman Sachs all recorded double-digit growth as of September. Margin debt reached $1.45 trillion in August, nearing the previous record of $1.5 trillion. 4
What are tax-aware hedging strategies? This quantitative research piece reveals the secrets.
Industry Dynamics
Quantiphi has recently launched campus recruitment. At an offline recruitment event in Nanjing, the firm offered high salaries, with intern daily wages starting at 2,000 RMB.
Founded in 2019, Quantiphi is a fund focused on secondary markets. The founder previously worked at global quantitative institutions like Two Sigma and Tower Research. With decades of industry experience, the team combines international research methodologies with local markets. The postdoctoral-background team possesses strong academic research capabilities and strategy translation skills, dedicated to creating sustained long-term value.
Quantitative Research
05 pyfolio-reloaded 0.10.0 Released: Multiple Numerical Fixes
Pyfolio-reloaded released version 0.10.0 on October 17, 451 days after the previous version 0.9.9.
This is a maintenance release focusing on numerical corrections, including: Sortino ratio downside risk calculations following the observation period, handling missing values in rolling regressions, positive clearing dates for short positions, slippage adjustments retaining zero P&L days, and trade inclusion at the start of live trading boundaries. These fixes may alter affected numerical results. Note: Portfolio relative round-trip statistics will throw an error if NAV (including cash) is missing; you must pass portfolio_value before calling. This version requires Python ≥3.11, supporting up to 3.15.
This library focuses on quantitative portfolio performance and risk analysis, compatible with Zipline. It integrates risk metrics and return visualization into “tear sheet” charts, with 616 GitHub stars.
Author Stefan Jansen is also the author of the Machine Learning for Trading series.
On the same day, he also released a new version (maintenance) of Alphalens-reloaded. Both libraries were previously developed by Quantopian; after Quantopian’s acquisition by Robinhood, Jansen took over maintenance responsibilities. 5
06 maverick-mcp v1.2.0 Released: Data Migration Required
MaverickMCP released v1.2.0 on October 17, 2026.
The toolset now includes 53 tools (38 core, 12 backtest, 3 research). Key note: Docker data is now defaulted to /data. You must back up the database under the old container’s /app before upgrading and mount named volumes. Backtests are now annualized based on 252 trading days; the profit/loss ratio for trades with no losses returns null; equity curves are truncated to 60 points. The PyPI name maverick-mcp-server is taken, so avoid installing from PyPI for now; use release tags or mirrors instead. Includes security fixes for soupsieve, among others.
This project is an open-source stock MCP server for local analysis, requiring no API Key to access Yahoo Finance. It supports market data, technical indicators, and SQLite portfolio management. It has 707 GitHub stars and is authored by wshobson. 6
07 How Much Did Duan Yongping Actually Buy?
Recently, Duan Yongping posted on Xueqiu that he “bought a bit” of a certain stock. It was later discovered that “a bit” meant 36.92 million shares. This attracted the attention of Zhihu user Bangbi Kuaipao.
Consequently, Bangbi Kuaipao created a knowledge base tracking Duan Yongping’s operations and freely released it on Zhihu AI Works under the project name “A-Duan Notes.”
A-Duan Notes correlates Duan Yongping’s “words” with his “actions,” allowing users to click back to the original text for each entry. The core collection includes 223 buy/sell and options operations, 41 companies, 31 quarterly 13F holdings, and 13,000 original posts, updated twice daily. Bangbi Kuaipao previously created a Buffett knowledge base and a Munger thinking models website; this is his third. 7
08 Decoded: What Are Tax-Aware Long-Short Strategies?
Tax-aware long-short strategies maintain market exposure close to the benchmark by financing long positions and shorting via securities lending. They leverage positions and achieve high turnover to harvest floating losses and defer floating gains, using capital losses to offset taxable gains and thereby enhancing after-tax returns.
AQR research indicates that their tax benefits primarily come from deferred gains rather than increased sales of losses.
Construction Path: Establish super-net exposure structures such as 130/30, 150/50, or 200/100. Use factor alpha models to overweight longs and underweight shorts, deviating around the benchmark. Write tax objectives into the rebalancing optimizer to continuously realize floating losses. Manage margin interest rates, borrow costs, and collateral haircuts; some exposure can be synthesized using index futures or box spreads. A 200/100 structure can harvest approximately 40% in losses in the first year, significantly higher than the ~15% for pure long direct indexation.
Risks include changes in financing and counterparty terms during stress periods. Tax considerations should be prioritized after risk, return, diversification, and cost.
Quantitative Career
This career section discusses something different: everyone should master some efficiency tools, which is part of a quant’s professional development.
This issue introduces a relatively new browser: Tabbit, from Meituan, available in international and domestic versions. The main difference is that the international version can use foreign large models, such as Astro, Opus 5.5, and Argon.
Unlike Chrome and other browsers, Tabbit deeply integrates web conversations, multi-step automated agents, intelligent tab management, and reusable “Skills” into the browsing experience. It can transform open tabs, PDFs, and local files into a unified personal working memory library.
This issue highlights a unique feature: /taste (Learn Mode). /taste is an AI memory and note-taking skill within the Tabbit browser ecosystem. Its core function is to automatically convert content you browse or watch into searchable, citable structured knowledge notes, serving as an MCP (Model Context Protocol) service for various AI tools.
When browsing the web, we sometimes encounter excellent websites, such as quantide.cn (😁, this article is published on quantide.cn). Some articles leave a deep impression, but reading them once is often insufficient for digestion.
In the past, you might use a “Read Later” extension or bookmark them to revisit in the future. In reality, you almost never look at them again, which is the origin of the internet meme, “Go to my bookmarks to gather dust.”
Tabbit brings a new approach. Bookmarking is not for appreciation, but for revisiting—or even remembering. This is first-principles thinking.
This is the function of /taste Learn Mode. It can directly convert the audio, video, or web pages you are browsing into interactive study and intensive reading tools:
- One-Click Multi-Format Generation: Supports 11 learning formats including flashcards, Cornell notes, charting, QEC (Question-Evidence-Conclusion) tables, and multiple-choice questions. For example, it can automatically organize a 9-minute history lecture into a categorized comparison table, or deconstruct a PubMed medical paper into a QEC logic table and export it as Anki cards.
- Precise Sourcing and Instant Testing/Evaluation: Flashcards generated from watching YouTube courses, etc., include timestamps. Clicking on a wrong answer jumps back to the corresponding video segment; reading CFA ethics codes or other web pages allows instant generation of mock exams with automatic grading, with explanations directly quoting original standards.
- Error Tracking and Reinforcement: When swiping cards or answering questions in the browser, the system automatically marks mastery levels and provides “Retry Only Wrong Answers/Review Due” functions to target weak knowledge points.
Make every browse contribute to your knowledge compound interest. This is why I recommend it to quants.
