Bulk deal data is published by NSE daily. A lot of the data comprises of clients buying large quantities and selling it during the day to benefit from small price changes. However, there are quite a few that are outright buy or sell. Since, these are high volume transactions, it makes sense to test whether there is a signal embedded in there.
TLDR: This data does have some signal broadly, but isn’t useful for going long for short term trades.
For raw bulk deal data from Jan 2012 – March 2023, click HERE.
This type of analysis can go in multiple different directions:
1. Naively checking if aggregate buys/sells of any stock, has any signal.
2. Observing if any specific client’s has any signal.
3. Aggregate volumes of buy or sell having any correlation with future returns.
4. The duration of applicability of signal, if any.
and, many others that you can imagine.
Here I will try to do the simplest analysis possible. I.e. No. 1 and 4 from the previous list.
Methodology:
1. Aggregate the bulk deal data.
2. Aggregate the CM daily bhavcopy data (cleaned if possible) for same dates.
3. Aggregate bulk deal data by date and symbol -> calculate net quantity -> partition into Buy or Sell based on net quantity.
4. Create matrix of returns of CM Data per stock for whichever duration holding period you choose (here it is 5, 20, 60, 120 days).
5. Do a left outer join of net quantity bulk deal data and CM returns data (left being bulk deal data).
6. Summarize and plot the return data.
Tip: NSE data has an anomalous data for date “13-AUG-20”. Remove that.
Those interested in syntax for the same in R language, click HERE.
The return data takes some time to calculate. If interested, the RData for the same is HERE.
Added Methodology:
The returns can be calculated in 3 ways:
1. Simplest (Naive): Using the day of signal’s Close price to whichever duration holding period (operationally wrong, since we buy the next day only) – NetNaive
2. Slightly correct with assumptions: Using the next day’s Close price as buy (assumption that you’re able to buy exactly at close) – Net
3. Most correct: Using next day’s price which is randomly selected between High and Low of next day (since buy can happen anytime) –NetRand
I’ve published the returns data and summary stats for all three methodologies in the following files:
1. BULK BUYS RETURNS RAW.
2. BULK BUY RETURNS STATS SUMMARY.
3. BULK SELL RETURNS RAW.
4. BULK SELL RETURNS STATS SUMMARY.
The returns are calculated for a Rs. 10000/- purchase of any stock which provided a buy or sell signal based on net quantity of bulk transaction on the previous day.
A graphical representation of the same would be useful for casual readers. Presenting only graphs for NetRand return methodology below.
BULK BUYS SIGNAL
BULK SELLS SIGNAL
Discussion:
As is seen from the data and graphs, going long after either Buy or Sell signal isn’t very useful.
However, the negative returns of 1 Week holding after a Buy signal seems to be a consistent signal to go short on. Since shorting overnight isn’t an option, maybe the stock FnO junkies would be interested in this finding.
Can’t write too much beyond this. Will try to keep doing more analysis and publishing.
Any questions? DONT COMMENT (I don’t have time to read them everyday), please ask on twitter @dhritiman_ch (Gamma.Pehelwan). I’ll do my best to reply.