Now, Faculty Focus with Scott Weisbenner. In this module, the focus is on Jake Wan Wong. Jake Wan Wong is an Assistant Professor of Finance at the University of Illinois, having moved here from National University of Singapore. Jake when teachers corporate finance for us and most of his research studies a behavior and influence of institutional investors. But Jake Wan also has a recent paper examining whether customer opinions expressed through Amazon ratings predict future firm performance. In fact, this is highlighted in July of 2016 in the Wall Street Journal. Rather than me talk about this interesting work, I had an epiphany. Why not ask Jake Wan himself to talk about it? Welcome, and thanks for joining us, Jake Wan. Thank you, Scott. I'm really glad to have the opportunity to talk on my research on Amazon reviews. Oh, I'm sure the pleasure is all ours and our listeners. In this module of the course, we've talked about inattention on the part of investors and how this could potentially lead to some predictability in stock returns. We also discuss some potential modern sources of information to get a sense of what are people thinking about, or at least searching for Google-. Yeah. Looking like Google Trend data. You're looking at products on Amazon, looking at the reviews and a particular change in reviews, and what that means maybe for future stock returns and profitability. What was your motivation? What was the idea behind this piece of research? So basically I had the idea when I was doing online shopping. Basically, it occurs to me that when I search for a product, I always look at the customer reviews, and basically, I thought if such information is valuable to me, does it also provide value routing information to the financial market? That's why I started to collect the data on Amazon reviews and to do an investigation of the investment value of customer reviews. The idea is that basically, if customer reviews provide new information to the financial markets, that it should predict subsequent stock returns, and also it should predict the fundamentals of the company. How did you get all this review data? It couldn't have been easy. Absolutely. I had some help from several RAs who were specialists, who are experts in web scraping, and they helped collect the data from amazon.com, and also it required other data work including to match the information from Amazon to publicly traded firms in the US. In all, it's a lot of data effort. Like reviews that I posted or my wife posted could be in your dataset? Absolutely. If it's for products produced by public companies. That's right. It'd be interesting to get a sense of what you found and I think to help us very prepared, you have some a figure and some tables that we could look at. Yes. I've had a few of tables and figures here. I can show you this figures. You always, it's great to illustrate something with like a key figure and right off the bat, something near and dear to a lot of people have to file taxes in the US is TurboTax, you have an interesting chart of Intuit and its stock price shortly after the TurboTax release. Yes, exactly. This concerns customer reviews of TurboTax 2002. It was released on November the first 2002 on amazon.com. On November 27th of 2002, customers started to flood amazon.com with native reviews about the product. Basically, consumers complained about an anti-piracy feature of the software. I see. The feature required that the costumer can only print tax returns and electronically file tax returns on the first computer on which the software was installed, not all of the computers that the customer is thought the software. So as you can see, following the first native review of the software, the stock price has dropped quite significantly from about $27 at the time of the first review to about $22, two months later. It seems to be quite as dramatic change in terms of the shareholder value for the company. Perhaps not surprisingly, the company lowered its earnings expectation for the fiscal year because of a weaker sales in March 2003. In this case, we see that customers has a whole seem to possess information about the company's cash flows as well as the subsequent stock returns of the company. I see. At least for this one anecdote, the negative turn in the Amazon reviews predicts future stock price changes as well as fundamentals with the earnings. But this is one anecdote. How do you go about systematically defining these abnormal customer ratings across all these many firms? Yes. In order to capture abnormal customer ratings where other words, with the surprises in customer ratings, I first compute the simple average star rating of all customer reviews posted for company's products in each month. This would be the average across all the products like across a given company like all the products for 3M for example. Exactly. I measure abnormal customer ratings as the difference between the average rating in a month and the average rating in the prior 12 months for the company's products. Okay. For example, if this company you aggregate across all the products the average rating is 2.5 for this month, over the past year, the ratings that average two then this would be like a positive 0.5, an uptick in the ratings for the products of that company. Exactly. Basically, a positive value in this measure indicates positive surprises and an active value indicates an active surprises. Just quickly, what could cause the economics behind why there'd be either an increase in this abnormal customer rating for a given firm or a decrease. What could cause that? I think there could be several reasons for the changes in abnormal customer rating. One is that the company introduces new products that are perceived differently as compared to his previous or existing products. There could be changes in the competitive landscape. For example, the Garmin GPS could be adversely affected when smartphones all have GPS available for free. Basically, this can induce a change in consumers preferences and tastes and it can also result in changes in abnormal customer ratings. Got it. Now we have this set for every firm, you have this measure of change and Amazon ratings across all their products. Let's get to the meat of the matter here. What are the results of terms of forecasting stock returns? Exactly. Basically, in each month I swore stocks into tercels, pretty cool. Thirds. Yes. The stocks in the lowest abnormal ratings will be the stocks that experience negative shocks to customer ratings. These are the T1 portfolio. Stocks that have experienced high abnormal ratings will be the T3 portfolio. These are stocks that have experienced positive shocks in customer ratings. If we look at the subsequent month stock return, we can see that T1 stocks experienced inactive but insignificant returns in a subsequent model whereas T3 stocks experienced positive and significant returns. This isn't where we're controlling for the risk here setting the benchmark in a four-factor model or controlling for market risk. The small first, the size factor, the value growth factor, and then also momentum. Exactly. Basically, after controlling for these risk factors we still find some Alphas, especially for the T3 portfolio. Right. This is good, non-trivial in terms of magnitude. If we just look at the stocks that have the biggest uptick in terms of the Amazon ranks for their products, this Alpha is 0.5 percent per month, that's a little over six percent on an annual basis. If you look at the final row which is like the hedge fund strategy long, the firms whose products ratings are going up short those that are going down this difference almost 0.8 percent on an annual basis almost 10 percentage points per year. It seems like there's actually some relevant information in these Amazon reviews in terms of forecasting future stock returns. Right. The economic marketing of synthesis requires significant here. You subjected this as like any serious study would do, you always want to look at where should this effect be larger, where should this effect be smaller? I understand you did that as well. Exactly. I look along three-dimensions as suggested by previous literature. In particular, I showed that the Alpha on the spread portfolio of high abnormal rating minus low abnormal rating seems to be a stronger among stocks that have high idiosyncratic volatility, low analyst coverage, and small cap firms. In the literature, it has been argued that these measures are proxies for limits to arbitrage and limits to investor attention. It seems to be consistent with the idea that limits to arbitrage and limits to investor attention drives some of this stock return predictability. The basic idea is high idiosyncratic volatility may be a lot of uncertainty about the firm low analyst coverage, people aren't following it that much small firms as well. Those might be the firms where you expect there to be more value relevant information contained in these Amazon ratings just because they're not known as well. Exactly. Yeah, so is will take some time for the information to be fully reflected in the stock price, resulting in this predictability pattern. Then key, I think to your paper here is then looking at, does this return effect that you show in the subsequent month, is this a price pressure effect that people are just reading about the rings, they're buying the stock, but the market's already incorporated this news, so you'd expect that price pressure effect to subside, the return effect to reverse, or is this a sustainable increase in price that you observe? What happens beyond this one month period? Yes. I looked also up to 12 months after the first month post formation of the portfolios, and I don't see reversal of this stock return predictability. Seems that the reviews convey information about the company rather than some non-information related factors. Interesting. Seems like you've presented some pretty compelling evidence that the Amazon reviews, or more specifically, the change in these Amazon reviews relative to the prior history predict stock returns and this effect on stock returns isn't reverse, so suggests like something fundamental is at play. Exactly. Do you have any evidence regarding predictability not just for stock prices but for future revenue, for future earnings in your papers? Yes. As part of the paper, I also show that there is a predictability of abnormal customer ratings for subsequent cashflow surprises. In particular, I look at revenue surprises as well as earning surprises. I show that WHEN abnormal customer rating is high, subsequently, the company is more likely to experience positive shocks to their revenue as well as positive shocks to their earnings. This is consistent with this cashflow stores. For the earnings this is already controlling for the best guess that the professional analysts estimates. These Amazon reviews can predict the earnings beyond what the experts are saying? Exactly. It seems to suggest that the average analyst hasn't incorporated all the information in customer reviews in his estimate, that still there is some predictability above the average analysts forecast. In this module, I presented some research that looked at searches for stock tickers on Google that seemed to be predictive of stock prices in the short-term, but then reversals in the long term. It's consistent with like a price pressure effect, like the searches on Google for stock tickers didn't seem to really reflect anything fundamental. You seem to have a different set of results. Why is that? Why the difference when you're looking at the Amazon reviews as opposed to the Google searches for stock tickers? That's a very good observation. I think the key difference here is that Google's search volume only captures the quantity of investors interest. It doesn't captures the direction of the underlying information. In particular, the high search volume does not necessarily indicate good news. It can be a buy or a sell. Right, exactly. One recent example would be the emission scandal at Volkswagen. I'm sure that when the scandal erupted, there was a lot of searches for the company, for VW. However, this indicates actually bad news for the company. On the other hand, the Amazon reviews provide us clear indication regarding the interaction of the underlying information. If the ratings are abnormally high, that it suggests good news for the company's cash flows whereas if it's negative, then it's bad news. The key thing with the Amazon ratings, you actually have this quantifiable measure. Is this good or is it bad? Is the rating five is at one? Exactly. Do sophisticated investors, to the extent to which you can study this hedge funds, do they anticipate this good news in the Amazon ratings? Or are they more seemed to react to the good news revealed by the Amazon? That's a very good question. Indeed look into the relationship between hedge fund rating and abnormal customer ratings. I don't find evidence that hedge fund can anticipate abnormal customer ratings. However, they do react to abnormal customer ratings in their decisions so basically I showed that following an increase in abnormal customer rating, hedge fund tend to buy more of the stock. You don't see hedge fund buying, let's say Microsoft right before there's great Amazon ratings for Microsoft products but you do see after the great ratings for Microsoft products on Amazon, hedge funds are more likely to buy? Exactly. This is from the 13 institutional filings, so you can see their long positions on the stock? Exactly. One final question here and you're looking at Amazon it seems to have this great wealth of information, some of which you've documented that predictable how that can be used to predict future firm performance. Should Amazon have a hedge fund or venture capital division based on all this great data they have on future trends in the economy? Yeah, that's a great idea so actually you were not the first one to think of that. Story of my life here. What would be the first time it happened? A article in Harvard Business Review actually commented the vast amount of data that Amazon has collected so basically they argue that Amazon is extraordinarily well-positioned to invest in specific accompanies and sectors because of the inflammation advantage that they may have. Well, this is really great stuff, Joaquin I think our viewers are going to see why we're so excited to hire you a few years ago in our department. Before we wrap up, I wonder if you'd be game for there's a favorite segment we have on the show called Awkward Moment so would you be up for it? Yes, that's good. Excellent, so let's take a quick break and then we'll come back for the awkward moment with Joaquin Wong. The end is definitely near for this interview with Professor Joaquin Wong for the faculty focus series for Module 4 and the course as well. But we're not done yet. Please see my upcoming videos, they will feature discussion of the economics of the mutual fund industry, as well as providing international perspective on the type of mutual funds and the fees associated with those funds from across the world and of course, you don't want to miss the course conclusion. Now more with Joaquin Wong. Joaquin, this is actually more of an awkward moment for me. Given all the Amazon packages that come to my house, it seems like we should probably have a distribution center and champagne. I was wondering if you could use your great data analytics ability and actually see if I give you the list of our purchase does not have any predictive power for anything except just to fall on our family's bank account? Yes. Well, that's an excellent question. My household has also contributed a lot to Avalon sales. I think there could be incremental information contained in the actual processes of consumers so it will be great if our viewers all send me their personal purchase records on Amazon so that I can infer the actions of the crowd to see whether there is incremental information contained in this aggregated actions by the crowd of consumers. Just a word of advice you might want to get IRB approval before you collect that data. But I'll be sure to at least give you our households data you could see if it predicts anything. Thanks so much Joaquin, this has a lot of fun. Thanks for joining us on our Faculty Focus segment. Thank you very much. It's a pleasure to be here. Excellent thanks and remember, in Module 4, we actually have two Faculty Focus segments. Please see my interview with Professor Josh Palette earlier in the module and as always, thanks for watching.
