Creating indices of TRS transactions on the fly for index forecasting
The US equity swap data (Total Return Swaps, or TRS) on Deriv1.com are producing multiple opportunities for analysis. One of them is the creation of NASDAQ 100 and S&P 500 long and short indices. We’ve set up a model where we run the numbers right when the TRS data go live, around 9AM. We’ve learned a few things so far.
As we show in the daily files available at https://deriv1.com/us-otc-equity-derivatives/, we split out what are evidently long and short transactions as well as the transactions that can’t be identified either way. Usually the non-identifiable trades are are around 25% of the last day’s volume, so we feel modestly confident that the long and short directional activity is telling us something, especially when the long/short split is more than 25% wide. We then aggregate the three different categories of volume for each ticker in the NASDAQ 100 and S&P 500 indices. This takes us around two minutes to do manually in Excel using a template that we’ve previously created. The result is a long/short/not identified percentage for each index. The same exercise can be performed for the entire market and for specialized baskets of interest.
We also have the weights of each index by security but find that the resulting long/short percentages aren’t that different than if we run the absolute values vs. if we weight the numbers. It’s probably worth running the calculation both ways to see the difference over a longer time horizon.
When we look at the resulting long/short split of the entire TRS portfolio and compare to the closing price of the next day’s S&P 500, we find a 68% correlation that a long bias on T+0 results in an up market outcome end of day on T+1. We haven’t run the full backtest on just the S&P 500 or NASDAQ 100 indices yet but see similar results on an anecdotal basis.
We also see some moves at the beginning of the T+1 day trading that make us think that TRS trades are happening at the end of the T+0 day that need to be hedged at market open. This results in some combination of higher futures prices or an elevated opening price on the main indices compared to the prior day’s close. More work remains to be done to be certain about the degree of correlation and what time frames it pertains to.
These experiments so far give us the impression that the TRS data are providing opportunities for institutional and retail traders to capitalize on this information flow. The creation of indices is helping show us how broad market movements are tracking the TRS data with a one day lag. This is a different exercise than identifying individual large outlier transactions, similar to what we found with Digital Realty Trust (DLR) last summer. Given the newness of the TRS dataset, we look forward to additional exploration and welcome feedback from data subscribers, including the opportunity to publish your own findings on this site.