“Anomalies are essentially patterns in stock return data that are not easily explained by standard risk models. In basic finance training, we are taught that financial markets are extremely efficient. So it is exciting to find parts that cannot be explained by risk and lead to alpha. I think that is why a lot of people are willing to spend time looking through data to discover ways that can help them beat the market. So if you discover an anomaly, it can give you an edge.”
“There is recent literature that discusses what actually causes anomalies. Out of the hundreds of reported anomalies, how many are genuine and how many are a manifestation of data mining? This is a huge debate in academic finance literature. For an anomaly to be a factor, it has to be very robust. There has to be a good reason for the existence of a premium. It should be reflected in the data over a long period and must still persist after it is publicized.”
“From an academic perspective, alternative datasets are really fascinating. They allow you to nowcast the fundamentals of a firm. We can find alternative data on parking lot capacity, satellite images or credit card usage, for example. Such datapoints can give you an idea about a firm’s fundamentals in real time. Company results are announced with a lag and cause delayed market reactions. So these alternative datasets can help predict future returns.”
“I am a little more hesitant about machine learning from an academic point of view. When we discuss data patterns, we want to understand what drives them. Machine learning techniques are still often ‘black boxes’. For example, a random forest algorithm can be really effective, but you might have no have insight into why certain criteria work. This can prevent you from having a good grasp of the fundamental economics. As an academic, I feel less comfortable with using a black box to improve our understanding of financial markets.”
If different inputs predict return in a very complicated way, then machine learning techniques can identify this quite quickly
“On the other hand, they are great tools for practitioners. They allow you to discover data patterns that would be otherwise difficult to pick up, especially in terms of non-linear relationships. So if different inputs predict return in a very complicated way, then machine learning techniques can identify this quite quickly. I know there are many hedge funds that use these techniques and are doing well. This is because they have tools that allow them to discover complicated and time-varying data patterns. So machine learning can actually work well in practice.”
“My research finds that there are potentially two drivers of short-term return reversals.1 One is compensation for providing liquidity. When large institutions try to sell huge amount of stocks in a hurry, it causes a stock price impact as they are pushed down temporarily. When the selling pressure disappears, the prices will typically recover. If you trade against this institution, you could benefit from this reversal as you could pick up the stocks at depressed prices and then wait for them to recover. Essentially, you will get compensated for providing liquidity.“
“Investor expectations can also drive short-term return reversals. For example, less sophisticated investors can extrapolate recent returns into the future. This can lead them to buy a stock that is trending up as they expect the good performance to continue, pushing the share price too high. When the market realizes that the firm’s fundamentals are not great as the share price suggests, then a correction or reversal can follow. To benefit from this, you could look for opportunities where this extrapolative behavior is present and results in excessive valuations, such as the meme stock phenomenon.”
“I am spending a lot of time looking at big data and investor behavior. For academic researchers, big data can unlock a gold mine of information and allow us to analyze investor behavior more closely. For instance, we are working on a series of papers that look at Bloomberg user activity. Bloomberg is like a social network. When you log into a terminal, your status is publicly available. So people know if you are logged in or not, or when you are active or idle. We are analyzing this type of information and can think of it as a measure of work effort.”
What really excites me going forward is being able to learn more about investor behavior by taking advantage of new alternative data sources
“Previously, we had no concrete idea of how many hours hedge fund managers or analysts worked for, as an example. Yes, there is survey-based evidence, but it is usually plagued by self-reporting bias. But with these alternative datasets, we are able to observe their activity much more closely. We can look into things like how the Covid pandemic affected their work habits, what motivates them to work longer, or if there actually is a positive relationship between working longer hours and performance. What really excites me going forward is being able to learn more about investor behavior by taking advantage of new alternative data sources.”
1 Da, Z., Liu, Q., and Schaumburg, E., March 2014, “A closer look at the short-term return reversal“, Management Science.; and Da Z, Huang, X., and Jin, L., April 2021, “Extrapolative beliefs in the cross-section: what can we learn from the crowds?” Journal of Financial Economics.
BY CLICKING ON “I AGREE”, I DECLARE I AM A WHOLESALE CLIENT AS DEFINED IN THE CORPORATIONS ACT 2001.
What is a Wholesale Client?
A person or entity is a “wholesale client” if they satisfy the requirements of section 761G of the Corporations Act.
This commonly includes a person or entity: