Just wondering - Why split the data into quantiles if only taking positions in the top 10? Would the pool of stocks not be large enough to take the top ten names (*liquidity dependent) without splitting into quantiles?
Also - The same names in biotech may be used for several days correct? Often large volume shocks are followed by multiple days of large volume
splitting the data into quantiles is the standard scientific way of researching these kinds of anomalies... but yes, once you are convinced your signal has predictive power, you can monetize it any way you like it (quantiles, deciles, quintiles, take the N top names, take the N top names with closing price above X, take the N top names with minimum turnover of Y, take the N top names of US-based companies only, etc)...
I haven't checked more days into the future. It's a valid research question/hypothesis (out of many many many new questions/hypotheses that can be tested in addition to this).
This is the idea: keep asking questions, testing sound hypotheses, and iterating. The more you test, the more likely you are to find a profitable signal :)
Great read—overnight returns have provided a persistent edge for years, and combining that with volume insights adds a powerful dimension. I recently published an article on a systematic volume-based signal for Bitcoin (https://alinakhay.com/p/a-systematic-volume-edge-for-spotting), and it’s fascinating to see how volume-driven signals can deliver predictive power across different markets. Really appreciate your thorough analysis—excellent work!
Just wondering - Why split the data into quantiles if only taking positions in the top 10? Would the pool of stocks not be large enough to take the top ten names (*liquidity dependent) without splitting into quantiles?
Also - The same names in biotech may be used for several days correct? Often large volume shocks are followed by multiple days of large volume
*Sorry new at any of this stuff
splitting the data into quantiles is the standard scientific way of researching these kinds of anomalies... but yes, once you are convinced your signal has predictive power, you can monetize it any way you like it (quantiles, deciles, quintiles, take the N top names, take the N top names with closing price above X, take the N top names with minimum turnover of Y, take the N top names of US-based companies only, etc)...
I haven't checked more days into the future. It's a valid research question/hypothesis (out of many many many new questions/hypotheses that can be tested in addition to this).
This is the idea: keep asking questions, testing sound hypotheses, and iterating. The more you test, the more likely you are to find a profitable signal :)
Do you use the Norgate Data to get the Tickers and use this to get the Intraday data from Polygon?
Correct
And do you use the real time data or 15-mins delayed from polygon?
why would someone use 15-min delayed data? :)
Just to save the money with the different abos
Maybe I’m missing it… what did you use to do the volume sorting?
Volume from the day* divided by exponential moving average of the volume (half life of 60 days, exactly as in the article).
*from 9:30 to 3:45 pm
Great read—overnight returns have provided a persistent edge for years, and combining that with volume insights adds a powerful dimension. I recently published an article on a systematic volume-based signal for Bitcoin (https://alinakhay.com/p/a-systematic-volume-edge-for-spotting), and it’s fascinating to see how volume-driven signals can deliver predictive power across different markets. Really appreciate your thorough analysis—excellent work!