causaLens Technology with 2iQ Data Predicts “Meme Stock” Behavior
The “meme stock” phenomenon that emerged in January 2021 grabbed headlines around the world, however many months prior to the start of that activity, there were signals present that predicted the dramatic trading frenzy.
Using high-resolution short interest data from 2iQ with causaLens technology, we previously predicted extreme price changes in GameStop as early as September 2020. Now with more data available, we extend our previous analysis from GameStop to AMC and other meme stocks.
Data
Two key quantities in 2iQ’s data are highly informative. Utilization rate is the percentage of shares available for shorting which have been shorted. To short a stock, short sellers borrow the stock and pay a loan fee: the loan rate provides insight into the demand for the stock on the short side. 2iQ provides further explanation of these features in this blog post.
We analyzed six instruments: four “meme” stocks (GameStop, Tesla, AMC, and Nokia) popular with Redditors, and two large stable companies (Google and Apple).
Analysis
We find that comparing the utilization rate with the loan rate of newly issued contracts is highly informative for predicting price movements. Examining the utilization-loan rate plot for January, we can see that GameStop and AMC were both in the upper right corner. This is a red flag for a short squeeze.
Fig. Utilization rate vs. average loan rate, log-transformed, for newly issued contracts in January 2021. Different regions of the plot exhibit different risk levels for short squeeze episodes. The bottom-left region is low risk, the center is medium risk and the top-right is high risk.
Fast forwarding to examine the same plot for May tells a very different story, while GameStop had moved towards the center of the graph, AMC stayed closer to the extreme end.
Fig. Utilization-loan rate plot for May 2021.
Let’s zoom in on the upper-right region of the utilization-loan rate plot, where the ‘Short Squeeze Risk Factor’ (SSRF) is high. For stocks in this region, we find that SSRF dynamics are highly predictive of price movements.
Take AMC. The absolute SSRF is high, a red flag. In the figure below we can see that the dramatic spikes in price occur following steep falls in the SSRF.
Fig. Short Squeeze Risk Factor vs. price for AMC, a high-risk stock.
Similarly for GameStop, the two big price jumps came from steep falls in SSRF. Using this index, it would have been clear that AMC was much more likely to experience a spike in May/June.
Fig. Short Squeeze Risk Factor vs. price for GameStop, another high-risk stock.
This behavior is not replicated by stocks occupying different SSRF risk bands. Take Alphabet and Apple stocks as paradigmatic low-risk stocks.
For Alphabet, we instead find that the SSRF simply tends to drift upwards with increasing price as investors become gradually more willing to bet that the asset is overpriced. But there is no clear indication that it is the SSRF which is driving the increase in price. (The same is true of Apple).
Detecting Causality in Short Interest Data
causaLens is designed to identify predictive causal drivers from time-series data. We have seen that the SSRF appears to have a significant influence on stock behavior. It is vital for effective modelling to determine whether the relationship between SSRF is a correlation or a true causation.
By using our panel dataset of large market cap companies, we find that the risk factor is identified as one of the top causal drivers alongside the average age of contracts and the number of units shorted.
Conclusion
In conclusion, the unprecedented behavior in meme stock prices can be more deeply understood by using Causal AI with high-resolution short interest data. In particular, the short squeeze risk factor is extremely useful for early identification of short squeeze events and at-risk assets.
2iQ closely partners with causaLens to go beyond conventional ML to deliver causal solutions for customers. More on the results of our partnership can be found here.