Detecting stuck quotes automatically

nitro · Feb 20, 2006

Quote from stephencrowley:

lol. That is definately a good idea. Doesn't help for historical quotes though.

Huh?

In your historical data, if the market is crossed or locked, assume the NBBO is bad and hit the next best tick in the book, assuming no uptick rule issues.

If your system is taking advantage of crossed and locked bids in stocks, good nite and gooood luck, especially sitting 45ms from the exchanges.

nitro

stephencrowley · Feb 20, 2006

No, I don't bother looking at locked/crossed markets.. I'm not that concerned about trying to scrape pennies like that.. but when playing back historical data you obviously can't execute it. So, back to the question, looking only at hostircal full depth quote data, how can you tell if the best quote is stuck?

I've thought about keeping a timestamp for each level in the book and some sort of monitor to check the time of the best relative to the next few levels.

Quote from nitro:

Huh?

If the market is crossed or locked, assume the NBBO is bad and hit the next best tick in the book, assuming no uptick rule issues.

If your system is taking advantage of crossed and locked bids in stocks, good nite and gooood luck.

nitro

nitro · Feb 20, 2006

Quote from stephencrowley:

No, I don't bother looking at locked/crossed markets.. I'm not that concerned about trying to scrape pennies like that.. but when playing back historical data you obviously can't execute it. So, back to the question, looking only at hostircal full depth quote data, how can you tell if the best quote is stuck?

I've thought about keeping a timestamp for each level in the book and some sort of monitor to check the time of the best relative to the next few levels.

Either I am confused or you are.

You have depth of book historical data. While replaying it, the Best Bid >= the Best Ask. If so, assume the top of book is bad and use the next best bid or ask in the book, etc.

nitro

stephencrowley · Feb 20, 2006

Right. Just remove that entry from the book and move on.. how to decide which quote is bad though, the bid or the ask? I suppose I could toss the best with the oldest timestamp. Actually.. that sounds like a great idea. Thanks for the help.

I guess if both quotes stick at the same time then I am just screwed.. that sounds pretty unlikely though.

Quote from nitro:

Either I am confused or you are.

You have depth of book historical data. While replaying it, the Best Bid >= the Best Ask. If so, assume the top of book is bad and use the next best bid or ask in the book, etc.

nitro

nitro · Feb 20, 2006

Quote from stephencrowley:

Right. Just remove that entry from the book and move on.. how to decide which quote is bad though, the bid or the ask? I suppose I could toss the best with the oldest timestamp. Actually.. that sounds like a great idea. Thanks for the help.

I guess if both quotes stick at the same time then I am just screwed.. that sounds pretty unlikely though.

That's one way of doing it. I would explore all ways and then watch when it happens in realtime to see which way is more accurate.

On the NYSE for example, the quote you see is not necessarily stuck in the technical sense - it could be that the spec or the asst does not have time to update it. There is no one right way to do it imo.

Your last possibilty can be handled by looking into the future and seeing which way it crosses first [in the historical data case]. Also, it is extremely rare because I have never seen it.

nitro

stephencrowley · Feb 20, 2006

Interesting.. I've been experimenting with combining prices from different quote sources as well.. although that gets more complicated, updating covariance matrices and whatnot.

Also, I rarely see stuck quotes on most of my feeds except ARCA, it sticks quite often for some reason... quite annoying.

Quote from nitro:

That's one way of doing it. I would explore all ways and then watch when it happens in realtime to see which way is more accurate.

On the NYSE for example, the quote you see is not necessarily stuck - it could be that the spec or the asst does not have time to update it. There is no one right way to do it imo.

Your last possibilty can be handled by looking into the future and seeing which way it crosses first [in the historical data case]. Also, it is extremely rare because I have never seen it.

nitro

nitro · Feb 20, 2006

Quote from stephencrowley:

Interesting.. I've been experimenting with combining prices from different quote sources as well.. although that gets more complicated, updating covariance matrices and whatnot.

Code:

x = Get NBBO
If x is not crossed or locked
   UpdateCovMat(x)
else {
   x = ExtractCorrectQuoteFromDOB
   UpdateCovMat(x)
}

nitro

stephencrowley · Feb 20, 2006

Cool. DOB is?

Also, my spread standard-deviation filter I posted above only works in high volume issues..

I implemented another filter which limits the max step size of the price in either direction..(for instance 1 penny per second) which might be bad in very fast moving markets, I could make the threshold adaptable based on short-term realized volatility I suppose. This however gets rid of all spurious outliers but doesnt help address the stuck feed problem. Two seperate issues.

nitro · Feb 20, 2006

Quote from stephencrowley:

Cool. DOB is?

DOB = DepthOfBook

Also, my spread standard-deviation filter I posted above only works in high volume issues..

I implemented another filter which limits the max step size of the price in either direction..(for instance 1 penny per second) which might be bad in very fast moving markets, I could make the threshold adaptable based on short-term realized volatility I suppose. This however gets rid of all spurious outliers but doesnt help address the stuck feed problem. Two seperate issues.

nitro