Last week, I published a post where we looked into 2.7 million retail 0DTE trades in order to find out what they were up to, and more importantly, how much they were making.
Long story short, spread sellers were doing fine, being positive on most days in the sample and making about $18 million more than option buyers.
Getting to that number meant rebuilding every multi-leg order that crossed the exchange, which left us sitting on a massive pile of data.
Now, we know that for a lot of option sellers, systematization is a core element of their strategy. For instance, “every day at time t, sell an n delta put”. So, if we know what they’re trading, could we also map out when they’re trading?
If so, then it naturally begs the question:
“If we know what they’re trading and when, how much can be made from simply getting in front of that flow ahead of time?”
Is this stuff even predictable?
Before putting a strategy together, we first have to figure out whether this flow is even as predictable as we think it is.
To do that, we can put together a simple v1 experiment:
For every completed trade, take its timestamp and measure how many seconds away it was from the nearest 5-minute mark
This allows us to see if any trades cluster at 10:00, 10:05, 10:10, etc.
Fingerprint each trade by its structure and direction
Short put spreads, short iron condors, long straddles, etc.
For each fingerprinted trade, count how many of its trades happen on the exact 5 minute wall-clock intervals
Split the sample days in half to test actual predictiveness
We’ll find the 10 most repeated results from days 1-10, then see if those same trades show up the same way on days 11-20
With that baseline test, we ran the experiment, and once again, the data pretty much speaks for itself.
Starting with every multi-leg order in the sample, the first finding is that the actual clock time is an extremely predictive factor: