Transcription

• Speaker #0

  Hello. Welcome back to Papers with Fact Test podcast. Today, we dive into another algo trading research paper. We're going to unpack abnormal trading volume and the cross-section of stock returns. Lee, Kim, and Kim from 2016.

• Speaker #1

  That's the one.

• Speaker #0

  Yeah. And the core question we're really tackling for you today is, you know, can unusual trading activity actually help predict future stock returns?

• Speaker #1

  And maybe what does it tell us about the market itself, right? Right. How it behaves.

• Speaker #0

  Exactly. So let's start with the basics. What kind of data were they working with here?

• Speaker #1

  Okay. So they really went broad. They looked at common stocks from the NYSE, Amex, and also Nasdaq. Right. And the timeframe was pretty extensive, January 68 all the way through December 2015.

• Speaker #0

  Wow. That's a lot of data.

• Speaker #1

  It is. And they build weekly and monthly return data, plus turnover data, you know, how often stocks traded. They did make some adjustments for Nasdaq volume, which is standard practice.

• Speaker #0

  Okay. And you mentioned abnormal volume in the title. How did they get at that? What was the key innovation?

• Speaker #1

  Yeah, this is really the core of it. Instead of just looking at like the total trading volume.

• Speaker #0

  The raw number.

• Speaker #1

  Exactly. They split it. They decompose it into two parts. Okay. First, there's what they call expected trading turnover or E-turn.

• Speaker #0

  E-turn. Got it.

• Speaker #1

  Think of this as the normal level, which you'd predict based on the stock's own trading history. It's usual behavior.

• Speaker #0

  So the baseline activity for that specific stock. Makes sense. What's the other part?

• Speaker #1

  The other part is the unexpected trading turnover or U-turn.

• Speaker #0

  U-turn.

• Speaker #1

  And that's basically, well, it's the leftover bit. Okay. The part of the volume that the past behavior doesn't explain.

• Speaker #0

  Ah, the surprise element, the abnormal bit.

• Speaker #1

  Precisely. It's the deviation. And understanding the split E-turn versus U-turn is fundamental to understanding your results.

• Speaker #0

  Okay, great setup. They've separated volume into expected and unexpected. Let's get right into the findings. With an E-turn, The expected part. Tell them about future returns.

• Speaker #1

  Well, this is where it gets interesting, maybe even a bit counterintuitive. They found that E-turn, whether you looked weekly or monthly, actually negatively predicted stock returns.

• Speaker #0

  Negatively. So higher expected trading meant lower returns later.

• Speaker #1

  That's what the data showed. Yeah. So stocks that everyone expected to trade a lot tended to underperform down the line.

• Speaker #0

  Huh. Any specific numbers on that? How strong was the effect?

• Speaker #1

  Sure. Let's take the monthly results. They sorted stocks into five groups, quintiles, based on E-turn. Okay. The group with the lowest expected turnover averaged an excess return of 0.95% per month. Right. But the group with the highest E-turn, that dropped to 0.54%. Wow,

• Speaker #0

  quite a difference, and weekly.

• Speaker #1

  Similar story, just smaller numbers. It went from 0.25% for the lowest E-turn group down to 0.19% for the highest. a clear downward trend.

• Speaker #0

  And was this just a quick thing or did it last?

• Speaker #1

  No, it seemed pretty persistent. The paper shows this negative effect lasting for up to 16 periods, weeks or months, depending on the analysis. Yeah. And they even mentioned in unreported analyses, it went out as far as 60 months, five years.

• Speaker #0

  So sustained high expected volume might be a bit of a, well, a longer term red flag, maybe?

• Speaker #1

  That seems to be the implication. And they check this against standard risk factors too.

• Speaker #0

  Ah, good. Like size. value, momentum.

• Speaker #1

  Exactly. They use the Fama French Carhartt four-factor model, standard stuff. Right. And even after adjusting for those known risk premiums, this negative prediction from E-turn, it stuck around. It didn't just disappear.

• Speaker #0

  Okay. So E-turn, negative predictor, persists long-term, robust to risk factors. Now, what about the other side? The unexpected part, U-turn.

• Speaker #1

  Ah, now U-turn. That tells a completely different story. A positive one, actually.

• Speaker #0

  Okay.

• Speaker #1

  At both weekly and monthly horizons, U-turn positively predicted future stock returns. So when trading activity was unexpectedly high.

• Speaker #0

  Returns tended to be higher afterward.

• Speaker #1

  In the short term, yes. That's the key difference. It's like this surprise volume carried positive information or maybe just attention that pushed prices up temporarily.

• Speaker #0

  That's a really neat contrast. Can you give us the numbers for U-turn like you did for E-turn?

• Speaker #1

  Absolutely. Monthly horizon, again, the lowest U-turn quintile had an average excess return of 0.45%.

• Speaker #0

  Okay. baseline.

• Speaker #1

  But the highest U-turn quintile, the stocks with the biggest positive volume surprises, they average 1.31%. Whoa,

• Speaker #0

  that's a big jump. Nearly a full percentage point difference per month.

• Speaker #1

  It is significant. And weekly was similar, going from 0.01% for the lowest U-turn up to 0.53% for the highest. Again, a strong positive trend.

• Speaker #0

  But you emphasized short term. How long did this positive effect last?

• Speaker #1

  Right. Not nearly as long as the E-turn effect. They found this positive predictive power for U-turn was really concentrated in the first few weeks, up to about five weeks.

• Speaker #0

  Only five weeks. What happened after that?

• Speaker #1

  Well, then things started to reverse quite significantly, actually.

• Speaker #0

  Reversals. So the gains faded.

• Speaker #1

  And then some. For the monthly portfolios, these reversals started kicking in around month four after formation. for weekly around week 14. And the reversal effect actually seemed to peak somewhere around the 10 to 12 month mark.

• Speaker #0

  So a short term pop from surprise volume followed by a longer term pullback. It sounds almost like an overreaction.

• Speaker #1

  That's definitely one way to interpret it. The market seems to react strongly, maybe too strongly to the unexpected activity initially, and then it corrects over time.

• Speaker #0

  Which leads us back to the whole point of splitting the volume,

• Speaker #1

  right? Exactly. Because if you just looked at the raw turnover. you'd get these confusing signals. Sometimes high volume looks good, sometimes bad, depends on the time frame. Right. But this decomposition helps explain it. The short-term predictability of raw volume, that seems to be mostly driven by the positive U-turn effect. Okay. While the longer-term picture for raw volume is probably muddied or even dragged down by that negative persistent E-turn effect.

• Speaker #0

  That makes a lot of sense. It clarifies why raw volume alone can be inconsistent. So speaking of raw volume, how did it actually perform in their tests when they didn't decompose it? Just total turn.

• Speaker #1

  Yeah, they looked at that too. At the weekly level, raw turnover to your end did show a positive relationship with next week's returns.

• Speaker #0

  Okay, short-term positive.

• Speaker #1

  Right. The highest turnover stocks did better than the lowest, and the difference was statistically significant. Q5 minus Q1 was positive. But my 45... Monthly was much weaker. Let's clear. They found sort of an inverted U shape, actually. The middle groups did okay, but the highest turnover group wasn't significantly better than the lowest. The Q5-Q1 spread wasn't significant.

• Speaker #0

  So again, the signal gets weaker or changes over slightly longer horizons if you just use the raw number.

• Speaker #1

  Precisely. It really highlights the value of that decomposition.

• Speaker #0

  Did they look at like a simple trading strategy based on raw turnover? Buy high, sell low?

• Speaker #1

  They did. And it basically confirmed this short-term versus longer-term story.

• Speaker #0

  How so?

• Speaker #1

  well A long, short strategy using raw turnover showed positive profits for like the first one or two weeks only.

• Speaker #0

  OK, very short term.

• Speaker #1

  Yeah. After that, the profits turned negative and actually stayed negative for quite a while, out to 16 months in their tests.

• Speaker #0

  Wow. So chasing high raw volume might work for a week or two, but then it seems to backfire.

• Speaker #1

  That's what the backtest suggests. It really underlines why just seeing high volume isn't enough. You need to ask, is this volume expected or unexpected?

• Speaker #0

  Yeah, that distinction seems crucial. Did they do other checks like... Making sure this U-turn effect wasn't just some other known anomaly in disguise?

• Speaker #1

  It did. They ran Fama-Macbeth regressions, which is a way to test if a factor predicts returns even when you control for other known predictors.

• Speaker #0

  Like size, value, momentum, liquidity?

• Speaker #1

  All those, yeah. Plus things like idiosyncratic volatility, analyst dispersion, earning surprises.

• Speaker #0

  And U-turn held up?

• Speaker #1

  It did. The positive predictive power of U-turn remained robust across different tests. Interestingly, U-turn's negative power was a bit less consistent in these shorter-term regression settings, but U-turn's positive short-term effect was strong.

• Speaker #0

  What about that idea of a high-volume premium from earlier research? Gervais, Kenyal, Mingle, Grin?

• Speaker #1

  Good point. They specifically checked that. They replicated the method from that 2001 paper to identify stocks that generally have high, normal, or low volume based on their own past history. Okay. And even within those groups. among stocks that are already known high volume traders, U-turn still positively predicted short term returns.

• Speaker #0

  So the U-turn effect, the unexpected part seems to be distinct from just being a generally high volume stock.

• Speaker #1

  That's the conclusion. Yeah. It's a separate phenomenon.

• Speaker #0

  Okay. So let's try to pull this together. The main takeaway seems to be that decomposing trading volume is really insightful.

• Speaker #1

  Absolutely key. Unexpected or abnormal volume. U-turn gives you a positive signal for short-term returns.

• Speaker #0

  But it reverses.

• Speaker #1

  But it reverses. Meanwhile, the expected volume E-turn is actually a negative signal, especially over the longer term.

• Speaker #0

  And this explains why raw volume gives mixed signals.

• Speaker #1

  Exactly. It resolves that inconsistency. The short term is U-turn's game. The long term is influenced more by E-turn's drag.

• Speaker #0

  That feels like the real aha moment from this paper. It takes something seemingly simple, volume, and shows there's important hidden information if you just look a bit closer.

• Speaker #1

  Definitely. And the paper... briefly touches on why this might happen behavioral stuff.

• Speaker #0

  Oh, like what?

• Speaker #1

  Things like investor overconfidence, maybe biased self-attribution after gains, the disposition effect, even just attention shifts. They suggest the U-turn effect might be more tied to these behavioral biases than just simple attention grabbing.

• Speaker #0

  Interesting. So maybe the unexpected volume reflects moments when these biases are driving trading?

• Speaker #1

  Plausibly, yes. They also found that short sale constraints didn't seem to make the predictability stronger, which sometimes happens with mispricing stories.

• Speaker #0

  OK, so wrapping up, the big insight is this decomposition.

• Speaker #1

  For sure.

• Speaker #0

  Which leads to a final thought, maybe for you listening. Even if you're not calculating E-turn and U-turn formally. Right. How can you use this concept? Maybe it's about paying attention to significant deviations from a stock's normal trading pattern.

• Speaker #1

  Yeah, perhaps looking for those unusual spikes or lulls as potential short-term signals.

• Speaker #0

  While being maybe a bit more wary of stocks that just trade heavily, consistently, month after month.

• Speaker #1

  Could be a practical way to think about applying this idea. recognizing that not all all volume is created equal.

• Speaker #0

  Definitely food for thought. This was a great breakdown.

• Speaker #1

  Yeah, really interesting how separating those components changes the picture so much.

• Speaker #0

  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. Hackey Trading.