5 Goals for Hedge Hiker in 2021
Time series & current holdings for 3000 hedge funds classified by strategy (e.g. L/S) & other data (e.g. turnover). Dynamic UI to create/save subsets of funds to analyze aggregate data/hunt for ideas.
I thought I’d send out a note to introduce this “Hedge Hiker” project and to share our 5 big milestones for the rest of 2021 so you know what’s coming.
If you haven’t subscribed yet, add your email below and I’ll send you updates once we have results to share.
Some Context on Hedge Funds
Over the last couple decades, hedge funds have returned about 250 basis points of alpha over the index before-fees - considering long domestic equities only, and not leverage, short positions, or alternative level 2 or 3 assets1.
Blind Cloning: We Can Do Better
Over the last two decades, “blindly cloning” consensus and/or conviction hedge fund picks in aggregate would
For those who “blindly clone” top hedge fund picks, a 45+ day lag eats into the edge a bit, but studies find they can still see marginal benefits. For a recent test, take a look at Goldman Sachs Hedge Industry VIP ETF (GVIP) since inception in December 2016. GVIP rebalances quarterly into stocks most frequently held as a top 10 position by hedge funds. Since inception in December 2016, GVIP has returned 21.7% annualized to the S&P’s 17.4%.
Not bad returns, but any strategy as simple as piling into the industry’s most crowded names will get arbitraged away. We can do better.
First, by considering signals from hedge funds in conjunction with fundamentals and/or technicals. This is about idea generation, not blind cloning. Along these lines, I’ll share research related to 13F replication, but that will be supplemental. Many of my members are in the HF industry and have your own insights on which hedge funds to track. For that reason, the goal of Hedge Hiker will be to build an interactive platform that users to access our database, create/save subsets of hedge funds, and hunt for ideas as they wish.
Second, by focusing on fund selection. We know that consensus and conviction signals are strongest among hedge funds that take a long-term, fundamentals-based approach to stock picking. Specifically, 13F’s are most useful for L/S, L Only, Market Neutral, and Event Driven strategies, particularly among funds with have high simulated active share and low simulated turnover. We put a great deal of emphasis on building the most comprehensive hedge fund database possible - including data on hedge fund strategy, size, active share, turnover, win rate by sector/industry, and a range of other quantitative/qualitative data.
Third, by focusing on signals. What signals correlate strongest to outperformance? Does this answer among different groupings of funds, ie by strategy, specialization, size?
Within the hedge fund industry, many of the most-widely respected voiced are emerging managers not yet widely-known outside of their circles. And a number of better-known funds have consistently beaten the market even as market environments have changed. One can only analyze so many businesses. Why not consider how the “smart money” is positioned when it comes to generating ideas?
Questions Hedge Hiker Will Seek to Answer
Some questions Hedge Hiker will seek to answer are…
What’s the most effective way to use 13F’s for idea generation?
What’s the signal and what’s the noise in 13F data?
Do consensus / conviction picks outperform?2
Are these signals any stronger among a subset of funds that are more fundamentals-based and long-term in their approach?
Using quantitative and/or qualitative considerations, how can hedge funds be grouped to create a stronger signal for specific areas of the market?
5 Milestones for 2021
I have 5 milestones for Hedge Hiker this year.
Finalize Hedge Fund Sample (in progress): Finalize sample and scrape current quarter holdings for each. 13F disclosures often have typos and data errors; I’ll clean them as best I can. Append sample with fund-level data that will help inform custom groupings later on (fund strategy, simulated turnover, simulated active share, fund size, etc.). Append key data on an individual security level (sector, industry, market cap, “style”, etc.).
Create Time Series (in progress): Scrape historical 13F data for the sample. Test what signals (consensus, etc.) work best for replication, among hedge funds overall as well as within individual strategies of funds. Previous research suggests that L/S, L Only, Event Driven, and Market Neutral strategies with low turnover and high active share are best for replication.
Append Time Series Data to Sample: Analyze time series and append meaningful fund-level data into the sample, such as simulated past performance (1-year, 3-year, etc.) and simulated “success rate” by sector, industry, “style”, geography, or market cap. Many hedge funds have reputations for expertise in particular areas. Would be neat to measure that.
Launch Website with Static UI: Launch website in beta with static user interface. Assign funds into subsets that will represent standard groupings in the initial website. Groupings will be based on fund strategies and other meaningful characteristics that aren’t mutually exclusive (Tiger Cubs, Buffett Acolytes, Micro-Sized Funds, Small Cap Focused, China Focused, Technology Focused, Value Oriented, GARP, etc.). User will select between pre-set groupings and the interface will display subset-level visualizations (subset consensus picks, subset performance over time, subset sector exposure over time, etc.).
Move to Dynamic UI (by late 2021): Transition website to a dynamic user interface that lets users create/save customizable subsets of hedge funds and displays subset-level visualizations.
The usefulness of this tool will depend on the quality of the dataset so I’m planning ample time for data collection, cleaning the data, and building out the dataset. The next round of 13F’s are due mid-August. My goal is to have a website up with a static UI (milestone 1-4) by then. If I hit that goal I’ll have a dynamic UI by the following 13F deadline in mid-November.
I think this thing could be as useful as anything out there so I’m pretty excited to get it built. Since announcing this project I’ve gotten feedback from a number of folks in the HF industry
No paid tier yet but I have a tip jar if you’d like to help me out.
Next time you hear from me will be in July when I have results to share.
My junior analyst Everett says hi. Pictured below sitting like a human.
Amir-Ghassemi, Faryan and Papanicolaou, Andrew and Perlow, Michael. “Aggregate Alpha in the Hedge Fund Industry: A Further Look at Best Ideas” (April 26, 2020). https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3586138
According to this Epsilon study, high conviction picks perform about equally to the rest of hedge fund portfolios. This study looked at all hedge funds from 1999-2018. This conclusion is consistent with research by AlphaClone, Meb Faber, and Dynamic Beta.
However, this Novus/Barclays study finds that high conviction picks “outperform” the rest of portfolios by 286 basis points annually. This study looked at 2004-2019. A key difference is that this study analyzes a subset of just L/S, L Only, Market Neutral, & Event-Driven funds, which they contend are those most likely to take a fundamentals-based and long-term view.
Both studies find that consensus picks earn statistically significant alpha vs the benchmark index.