Hello, welcome back to Papers with Backtest podcast. Today we dive into another algo trading research paper. Yes, this one's called Alpha Cloning Following 13F Fillings. The authors are Randy Cohen, Christopher Polk, and Bernhard Sille. And the draft we're looking at is from March 18, 2009. It's a really interesting premise, basically asking, can we look at what mutual fund managers report holding those 13F filings and actually find profitable ideas? Specifically...
They're like high conviction ideas. The paper calls them best ideas. Do those best ideas actually outperform the market? That's the core question. And for you listening, the potential payoff here is maybe a shortcut, a way to identify strategies that might generate alpha. Right. So our mission today is really to dig into the trading rules they tested and crucially the backtest results. How did these best ideas actually perform? OK, so let's start there. How did they even define a best idea?
It can't just be the biggest holding, right? No, that's the key insight. They figured a manager's real conviction shows up when they overweight a stock compared to some baseline, some benchmark. Overweighting. So holding more of it than you'd kind of expect. Precisely. Like way more than its simple market cap might suggest, or more than other stocks they hold. It signals they really believe in it. Makes sense. And they didn't just use one benchmark.
They had four different ways to measure this overweighting. or tilt as they call it. Yeah, four specific measures. The first one they called market tilt. Okay, market tilt. How does that work? It's pretty simple, actually. You just take the stock's weight in the fund's portfolio and subtract the stock's weight in the overall market portfolio. So market tilt, T-E-T equals I-T-I-M-P-O-O. A bigger number means more overweighting versus the market. Got it.
Straightforward comparison to the whole market. What's the second tilt measure? The second is CAPM tilt. This one adds a layer. It takes that market tilt we just talked about and scales it by the stock's idiosyncratic variance. Whoa, OK. Idiosyncratic variance. Yeah, basically how much the stock's price moves around for reasons other than just following the market. They estimate it using like 60 days of return data.
So the idea is if a manager overweights a stock that doesn't just track the market, that's an even stronger signal of conviction. That's the thinking. The formula is CAPM TILTY PSI. That sigma squared term is the idiosyncratic variance. Okay, that makes intuitive sense. A bolder bet. What were the other two? The third is Portfolio Tilt. Here, the benchmark changes.
Instead of the whole market, you compare the stock's weight in the fund to its weight in evaluated portfolio of all the stocks that specific manager holds. Ah, so comparing it to their other choices. Exactly. It's Portfolio TILTY IVA FEE. Are they leaning into this stock even compared to their own usual picks? Interesting. And the fourth one, I'm guessing it combines ideas. You got it. CAPM portfolio tilt. It's that portfolio tilt. Scaled again by the stock's idiosyncratic variance.
So key APM portfolio tilty tilt equals it IPAA fee. Right. Applying that independent movement factor to the comparison within their own holdings. Yeah. And for all four, a higher tilt value suggests stronger belief from the manager. Okay, so we have these four ways to spot potential best ideas. What data did they use to test this, to see if it actually worked? They used standard sources, pretty much. CRSP for the stock return data. Right, the usual suspect for academic work.
And Thomson Reuters for the mutual fund holdings data, specifically the 13F filings. And the time period. It was a decent chunk of time, January 1991 through December 2005. Okay, and they looked at specific types of funds. Yeah. U.S. domestic equity funds. They filtered them a bit, had to have at least 20 stocks, over $5 million in assets. And importantly, they excluded index funds and tax managed funds, wanted active managers. Makes sense. All right. Definitions, data.
Yeah. Let's get to the results. What happened when they built a portfolio of these best ideas? Did it work? This is where it gets pretty compelling, I think. They created an equal weighted portfolio of all the identified best ideas from all the managers in their sample.
rebalanced quarterly and they found positive average monthly excess returns that's return returns above the risk-free rate across all four tilt measures we're talking ranges like 1.26 percent to 1.88 percent per month on average okay that's substantial per month per month but of course the next question is risk right where they just loading up on risky stuff exactly my thought Did they adjust for risk? They did.
They ran factor regressions using the standard Carhartt four-factor model, you know, market size value momentum. Uh-huh. The Fama French plus momentum. Right. Even after adjusting for those factors, the portfolio still showed positive alpha. Depending on the tilt measure, it was around, let's see, 0.29% to 0.78% per month, statistically significant in most cases. So still outperforming even after accounting for those common risk factors. That's impressive.
Yeah. They also used a six-factor model, didn't they? Yes, they added two more factors, one for idiosyncratic volatility, that's stock-specific risk again, and one for short-term reversal. And what happened with six factors? Did the alpha disappear? Actually, no. In many cases, the six factor alphas were even higher and more statistically significant, ranging from about 0.39 percent up to 1.15 percent per month. Higher alpha with more risk factors. How does that work?
Well, it suggests the standard factors weren't fully capturing the risks or characteristics of these stocks. These added factors might be relevant. Interesting. And did any tilt measures stand out with the six factor model? Yeah, the two portfolio based tilts. portfolio tilt and CAPM portfolio tilt, the ones comparing a stock to the manager's own holdings. They generally show the strongest results, higher alphas and T-stats.
OK, so maybe looking at conviction relative to their other picks is the most powerful signal. Now, what about these best fresh ideas? What was that about? Ah, yes, that was another layer. Best fresh ideas were defined as those best ideas where the fund manager had actually increased their position during the most recent quarter. So not just holding a big position, but actively buying more. Exactly. The thinking is... that's an even stronger, more current signal of conviction.
And did that pan out in the results? It really did. Table 2, panel B shows this. The risk-adjusted returns, especially those six-factor alphas, were consistently better for the best fresh ideas. They ranged from 0.46% up to about 1.27% per month. So the act of buying recently adds another layer of predictive power. That's a key potential trading role right there. Definitely seems like it. Focus on the recent buys among the high conviction holdings.
Did they look at like how high the conviction needed to be? Does it pay to focus only on the absolute top bets? They did analyze that in table five. They broke down performance by different thresholds, the top 100 percent of tilts, top 50 percent, and then really zooming in on just the top five percent. Wait, did the performance get stronger at the very top? Yes, significantly stronger. The outperformance wasn't just present. It was. amplified when focusing on the most extreme tilts.
Any standout numbers there? Oh, yeah. The most striping one, I think, was for the top 5% of the CAPM portfolio tilts. Remember, that's high conviction relative to their own holdings adjusted for idiosyncratic risk. That portfolio generated a six-factor alpha of 1.88% per month. Get out. 1.88% per month. That's huge. It is. Annualized, that's over 22%. Now obviously that's a back test, but it suggests that isolating those really, really high conviction bets could be extremely powerful.
That feels like a major finding. Forget diversification. Maybe it's about finding those few killer ideas. They also tested a best minus rest strategy. What was the point of that? Right. Table seven. That was a neat test. The idea was to go long the manager's single best idea, identified by one of the tilt measures, and simultaneously short the rest of that manager's portfolio, weighted proportionally.
So directly betting that the best idea outperforms their other holdings, trying to isolate the skill and picking that top stock from their general style. Exactly. Controls for manager-specific effects, sector bets, that sort of thing. And did that work? Did the best idea actually beat the rest? It did. They found statistically significant positive six-factor alphas for this strategy across all four tilt measures.
Really reinforces the point that there's something special about those top picks, doesn't it? It certainly seems to. It's not just that the manager is good overall. Their best idea genuinely seems better than their other ideas. Okay, so the single best idea seems potent. What if they included, say, the top three or top five ideas instead of just the single best? Did they look at that? They did in Table 8. They expanded the long side of that best minus rest strategy.
So long top three, short the rest. Long top five, short the rest. Yep. And interestingly, what they generally found was that as you included more stocks on the long side, going from top one to top three to top five, the resulting alpha tended to decrease. Ah, so diluting it with slightly less best ideas watered down the effect. Seems that way. It supports the idea that the conviction signal is strongest or maybe most valuable for that very top pick. Fascinating.
Okay, moving beyond the core performance, they also looked at characteristics of the stocks themselves, right? Like liquidity. Yes. Table 9 looks at liquidity using the bid-ask spread as a proxy. They split the best ideas into high liquidity and low liquidity groups. And what did they find? Does liquidity matter for cloning these ideas? It seemed to matter quite a bit. The finding was that the less liquid, best ideas generated the vast majority of the alpha. Really? So the harder to trade ones?
Yeah. The more liquid, best ideas, the ones easy for everyone to trade, showed much weaker results, sometimes even negative alpha. Huh. So maybe the edge comes from ideas where information isn't instantly priced in, or where it's costly for others to trade. That's a plausible interpretation. It suggests you might want to look beyond the big, obvious, highly liquid names if you're trying this. Okay, less liquid. What about popularity?
Did it matter if lots of other managers also flagged the same stock as a best idea? That's in table 10. They measured popularity by basically summing up the tilt ranks for a stock across all the managers. A stock that many managers tilt heavily towards is popular. Right. And the result? Better to follow the crowd or go against it? The result suggested going against the crowd, actually. The vast majority of the abnormal returns came from the least popular best ideas.
So the high conviction bets that aren't on everyone else's radar. Exactly. It hints that maybe the contrarian best ideas are the ones with the most potential alpha left in them. Interesting. So less liquid and less popular seems to be the sweet spot. Did they look at the funds themselves? Like, does it matter if the fund is huge or tiny or really concentrated? They did briefly touch on that in tables 11, 12 and 13.
They looked at fund concentration using the Herfindahl index, focus based on the number of holdings and just fund size assets under management. Any strong conclusions there? The results weren't always uniformly statistically significant, but there was definitely a trend. OK. What was the trend? It seemed that best ideas coming from smaller funds and funds that were more concentrated, holding fewer stocks overall, tended to perform better.
Maybe smaller, more focused managers have a by handle on their picks or less pressure to over diversify. That could be part of it. They might be nimbler or perhaps their best ideas haven't been diluted quite as much by holding hundreds of other stocks. Okay. That's a lot of ground covered. If we boil it down, what are the main takeaways for you, the listener, maybe thinking about this alpha cloning idea?
Well, first, it seems managers do have some ability to pick stocks that outperform, at least their top conviction ones. Right. The best ideas defined by that overweighting show real alpha. Second, focusing on best fresh ideas, the ones they recently bought more of, might give you an extra edge. Uh-huh. Recency matters. Third, But the less liquid and less popular best ideas seem to be where most of the outperformance is found. Don't just chase the obvious names. Go where it's less crowded, maybe.
Seems like it. And fourth, there's maybe a slight advantage to looking at the picks from smaller, more concentrated funds. And the big picture implication seems to be that maybe the reason many funds underperform overall is because they diversify too much, watering down the impact of their genuinely good ideas. That's certainly what the paper suggests. Their best insights get lost in a sea of other holdings. So lots of potential rules and ideas there. It definitely leaves you thinking.
It really does. And it poses a provocative question for you listening. If this research holds up, how could you actually use public 13F filings to implement some version of this? What are the practical hurdles, the data needs, the timing issues you'd face trying to become an alpha cloner? Yeah, bridging that gap from research paper to real-world trading strategy. That's the challenge. A great thought to end on. Thank you for tuning in to Papers with Backtest podcast.
We hope today's episode gave you useful insights. Join us next time as we break down more research. And for more papers and backtests, find us at https.paperswithbacktest.com. Happy trading.