Hello, welcome back to Papers with Backtest podcast. Today, we dive into another algo trading research paper. Today, we're digging into analyst coverage, information and bubbles by Andrade, Bian and Birch from the Journal of Financial and Quantitative Analysis published back in 2013. Right. And this one tackles a really interesting question. What role did analysts actually play in that pretty wild 2007 stock market bubble over in China? Exactly. And, you know, more importantly for our listeners.
What can we learn from that for maybe building trading rules or thinking about backtesting around those kinds of, well, extreme market conditions? Yeah, because the 2007 Chinese market was quite unique, wasn't it? Oh, absolutely. You had the government itself warning about a bubble forming. Really? The government? Yeah, and tons of new retail investors just piling in, plus major restrictions on short selling. Okay, so limited ways to bet against the market. Precisely.
And it seems a lot of these new investors were actively looking, you know, seeking out analyst reports. probably trying to figure out what to do. Hmm. Interesting. So they weren't just passive observers. And the sort of headline finding we're going to explore is that SOX, with more analysts covering them, actually had smaller bubbles. That's the core finding, yes. Counterintuitive, perhaps, but that's what they found. Okay, so let's unpack this.
When we say bubble here, what did that actually look like in the data for 2007 China? Well, the paper really paints a vivid picture.
You saw P-E ratios just... skyrocketing trading activity turnover went through the roof huge volumes massive and loads of new retail trading accounts being open even things like google searches in china for stock market terms just spiked really textbook stuff sounds like classic exuberance across the board did they focus on a specific period for this bubble analysis they did they primarily looked at November 29th, 2006 through to May 29th, 2007. Okay, about six months. Right.
And that end date is key because it was just before the government unexpectedly tripled the stamp duty, the security transaction tax. Oh, the tax hike. That must have been a shock. It definitely seemed to prick the bubble. Yeah. It was a major turning point. Got it. So a period of, let's say, rapid inflation and then a specific catalyst. Now, how did they measure the bubble intensity for individual stocks? Must have been tricky. Yeah, they didn't just look at the whole market.
They used five different measures at the stock level. First was just the cumulative return over that six-month window. Okay, how much did stocks go up? On average, around 200%. They tripled. Wow, tripled in six months. Yep. So you can imagine if you were backtesting a simple trend-following strategy then, the returns would look amazing. Phenomenal, but maybe misleading given the context. Yeah. Screams bubble risk, right? Absolutely. Big asterisk, as you say.
Second measure was the average P.E. ratio during that period. Price to earnings. Right. And they found very high average and median P.E. ratios. Clear signal of potential overvaluation. Okay. So value strategies would likely be flagging these stocks as expensive. Definitely. A flashing red light for value investors. Third was the announcement return. What's that? It's the return after that transaction tax hike was announced. Specifically, the five-day return.
Ah. So how stocks reacted to the bad news. Exactly. The thinking is stocks inflated by speculation would get hit harder when trading suddenly became more expensive. Makes sense. Less reason to flip them quickly. And that's what they saw. Significant negative returns on average for stocks after the announcement. Which also tells you something important for backtesting, right? Transaction costs matter, especially sudden changes. Absolutely.
Your models need to account for that possibility, especially around policy shifts. OK, what else? That's three measures. Fourth was a composite bubble measure. Basically, they combine the first three metrics return, PE, announcement reaction, into a single score for each stock's bubbliness. Kind of overall indicator. Yeah, just a holistic view. And fifth, they looked at the China HK premium. The price difference for stocks listed in both places. Right.
For companies listed on both the mainland, Shanghai Shenzhen and Hong Kong exchanges. Since Hong Kong was more open, had short selling. The difference could reflect the mainland bubble. That was the idea, though the short selling limits in China made pure arbitrage tricky. Still, it's another data point on relative valuation. OK, so five different ways to gauge the bubble intensity per stock. Do they all point the same way? Pretty much.
Yeah. The paper shows they were all significantly correlated, gave a consistent picture of which stocks were, let's say, frothier. Got it. So we have the bubble. We have ways to measure it. Now, the main event, analyst coverage. How did they measure that and what was the connection? They used a pretty straightforward measure, just the number of cell site analysts covering a specific stock. That was their proxy for how much public information was being produced about it. Simple count.
And the key finding, the really interesting part, was a strong negative correlation between the number of analysts and all five of those bubble intensity measures. So, weight more analysts meant a smaller bubble. Less froth. Exactly. Stocks with greater analyst coverage experience smaller price run-ups, lower PEs during the period and less negative reaction to the tax hike. That's quite something.
So the implication for, say, backtesting trading rules around bubbles, maybe filtering by analyst coverage could identify less risky assets. That's definitely a potential insight. It suggests that stocks under more scrutiny might have had less extreme behavior. More information flow seemed to dampen the speculative excess. Did they check if it was just a size effect? You know, bigger companies get more coverage and are less bubbly anyway? They did.
They controlled for market capitalization and other factors, and the relationship held. It wasn't solely a large-cap phenomenon. Analyst attention itself seemed to matter. Huh. That's important. Okay, but what about when analysts don't agree? That must muddy the waters. It absolutely does. They looked into that too, measuring the dispersion or disagreement among analysts. based on their earnings forecasts and their buy-sell recommendations. So how spread out were their opinions? Right.
And they found that the bubble-mitigating effect of analyst coverage was weaker when there was more disagreement among them. Ah, okay. So just counting analysts isn't enough, if they're all saying different things. Then the coverage doesn't seem to have the same stabilizing effect. The lack of consensus maybe reduces the coordinating power of the information. That makes intuitive sense. For trading rules, then...
High disagreement could be another red flag, maybe signaling more uncertainty, even if a stock looks well covered. Precisely. It suggests you might need to look beyond just the raw number of analysts and consider the consensus or lack thereof. Interesting. Did they also look at trading activity like turnover? How did that relate? Yes, they did. They found that more analyst coverage was generally associated with lower stock turnover. Lower trading. Why would that be?
The interpretation is that maybe more available information leads to more informed, perhaps buy and hold type decisions and less speculative churning or flipping of shares. More conviction, less noise trading. Possibly. But again, mirroring the other findings, this turnover reducing effective coverage was also weaker when analyst disagreement was high. OK, so another potential signal for backtesting. During bubbly times, high turnover combined with low coverage or high disagreement.
might flag more speculative, riskier stocks. That seems like a very reasonable takeaway, yes. Now, the paper also digs into why coverage might have this dampening effect. The obvious first guess is... That analysts were warning everyone, putting out cell ratings, the voice of reason. You'd think so. But surprisingly, when they analyzed the actual recommendations over time, they didn't see a big shift towards cell ratings around the bubble's peak. No. They weren't getting more bearish.
If anything, the average recommendation became slightly more optimistic, not less. Huh. So it wasn't analysts acting as a reality check, explicitly telling people things were overpriced. Doesn't seem like that was the main driver, no, which kind of weakens the idea that they were just reducing over-optimism directly. So if it wasn't explicit warnings, what's the alternative explanation for why more coverage meant smaller bubbles? The authors leaned towards a belief coordination mechanism.
The idea is that having more information out there being discussed and analyzed. Even if it wasn't uniformly negative. Right. It helped investors converge towards more similar, maybe more anchored views of a stock's fundamental value. It reduced the scope for wildly divergent expectations that can fuel a speculative frenzy. Ah, I see. So it's less about the content buy-sell and more about the process of information dissemination leading to more aligned beliefs.
That seems to be their main argument. More shared information reduces the heterogeneity of beliefs that bubbles often thrive on. That's a more subtle mechanism. Now, the big question with studies like this is always cause and effect, right? Endogeneity. Could less bubbly stocks just attract more analysts? A valid concern always. They tackled this quite carefully. First, they used lag analyst coverage from 2005, well before the main bubble period kicked off.
And they still found that 2005 coverage negatively predicted the bubble intensity in 2007. That helps mitigate reverse causality concerns. Okay, using past coverage predicts future bubble size. Yeah. What else? They also used instrumental variables, things that likely affect analyst coverage, but probably not bubble intensity directly except through coverage. What kind of variables? They used trading volume and mutual fund ownership, again from 2005.
Using these instruments, they still found evidence supporting a causal link from coverage to smaller bubbles. Right, using IV techniques adds more weight. Did they consider other stories? Like maybe analysts are just lazy and cover safer stocks, or perhaps it's about pump and dump schemes in the low coverage stocks.
They did address those alternative explanations, lazy analysts, institutional selling patterns, pump and dump, and argued that the beta didn't really line up strongly with those being the primary drivers. OK, so the core findings seems pretty robust. Let's try to synthesize the key insights for someone, you know, building trading rules or running back tests today, especially if they're looking at potentially frothy markets. Sure. I think there are several practical points.
First, if you see conditions resembling a bubble, rapid price run ups. High volume may be considered analyst coverage as a potential risk filter. So lower coverage might signal higher bubble susceptibility. Could be a reason to underweight or avoid. Potentially, yes. Second, don't stop at just the number of analysts. Look at the dispersion in their forecasts or recommendations. Because high disagreement might negate the stabilizing effect. Exactly.
High coverage with low disagreement might be the most stabilizing combination. High disagreement could still signal trouble. Got it. What else? Third, pay attention to turnover. High turnover in stocks with low coverage or high disagreement during these periods could be a warning sign of excess speculation. Another potential risk flag. Fourth, that transaction tax hike reaction is a stark reminder. Bubble prone assets can be very sensitive to trading costs and policy changes.
So robust transaction cost modeling and backtests is crucial, especially anticipating potential shifts. Definitely. And finally, while the China HK premium was hard to trade directly due to restrictions there. It hints at potential valuation gaps. Right. It highlights that cross-market discrepancies can exist, especially in bubbles, and might offer opportunities in less restricted markets or if new instruments become available. OK, that's a lot of food for thought.
So the big picture takeaway seems to be that public information proxied here by analysts coverage wasn't just noise during the 2007 China bubble. It actually seemed to play a moderating role. Yeah, likely by helping coordinate investor beliefs, even if the analysts themselves weren't explicitly bearish. It suggests information flow can be a subtle but important factor in market stability or instability. Fascinating stuff.
Definitely relevant for thinking about risk and potential signals when navigating potentially overheated markets in our backtests and strategies. It really underscores how complex market dynamics can be during these extreme periods. Thank you for tuning into 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.