Anyone have a SPY algo running intraday?
- Thread starter Fireplace
- Start date
yup
Inaccuracies with OHLC and volume. If I compare, say a 4 min bar with a corresponding 240 second bar, Esignal usually shows differences. If I comb through the 1 sec chart, the 4 min bar will not include some crazy prints while the 240 sec bar will, skewing the data.
If you use esignal, go to the 1 sec chart for the SPY and scroll through one one. You'll most likely find a ton of prints that were way outside the inside market, yet are counted in the tick interval data.
If you use esignal, go to the 1 sec chart for the SPY and scroll through one one. You'll most likely find a ton of prints that were way outside the inside market, yet are counted in the tick interval data.
I've done both, although I do not trade off of 1 sec data but have seen the chart build in real time and have encountered the same issues as I find if I just go back to previous days data.
I usually go through each day after the market closes to confirm my algo concept based on the new day's data and have also used historical data to also confirm if my idea works. I just know that I'm going to need a much more accurate tick interval data than I currently have when I decide to have this algo concept coded (in the not too distant future, hopefully).
I usually go through each day after the market closes to confirm my algo concept based on the new day's data and have also used historical data to also confirm if my idea works. I just know that I'm going to need a much more accurate tick interval data than I currently have when I decide to have this algo concept coded (in the not too distant future, hopefully).
From memory, eSignal doesn't filter tick data; the tick data you get (either in real-time, or also if you download historical tick data) is the datastream they compile in realtime from their various sources, i.e. it includes "bad ticks" (which are often trades reported late, and so out of sequence).
Realtime bars have the same "bad ticks", but historical bars get cleaned up.
NB If your strategy is going to trade in realtime, it's going to have realtime (i.e. "dirty") data, not the "clean" historical stuff...
... but still, for backtesting purposes, there can be a valid argument for "cleaning" up the data before you test with it; i.e. to increase the signal-to-noise ratio so that you can try out your ideas in a more "tame" ideal setting.
My guess is that any tick data you get that is cleaner than eSignal's has been cleaned up by the provider. And it is worth finding out how they have done that, so that you know how far they have departed from reality. Or clean the data yourself, so that you know exactly what's been done. Or use someone like this guy ... tickmaster.webs.com
Realtime bars have the same "bad ticks", but historical bars get cleaned up.
NB If your strategy is going to trade in realtime, it's going to have realtime (i.e. "dirty") data, not the "clean" historical stuff...
... but still, for backtesting purposes, there can be a valid argument for "cleaning" up the data before you test with it; i.e. to increase the signal-to-noise ratio so that you can try out your ideas in a more "tame" ideal setting.
My guess is that any tick data you get that is cleaner than eSignal's has been cleaned up by the provider. And it is worth finding out how they have done that, so that you know how far they have departed from reality. Or clean the data yourself, so that you know exactly what's been done. Or use someone like this guy ... tickmaster.webs.com
There are not any platforms that are "retail" based that can do that to my knowledge. Most can do tick or minute based but cannot create bars from X seconds etc.
There are platforms that can do this but they are very expensive to the point of being cost prohibitive for anyone but institutions. (I am talking in terms of 1000s to 10s of 1000s per month)
Even tick based bars get tough because of the amount of data required to build them. Most platforms limit the number of bars and the amount of tick data a trader can have access to.
Once you delve into this area, the next logical "want" is to be able to have the program change the calculation time frame on the fly - for example we want to compute the moving average on a 50 tick chart but only if the 334 second chart is in an uptrend.
Or you might want to compute something on 100,200,300 seconds and then only use the one that has been tracking the best (least error) the last 5 days for all calculations the rest of the day in the indicator or strategy.
Or you want to be able to code something and then find the best "time frame" to trade it on - which requires optimizing time.
The list goes on and on the ideas you can test once you have this ability. If anyone knows of any platform that can do this for an average trader/developer I would love to hear about it.
Multicharts can build second charts in whatever granularity you want.
From what I have seen when looking at the same datafeed it can vary slightly when you compare tick data (from which second charts are build) to the same feeds minute or daily data. For example the day high could be one or two ticks off - stuff like that. But this happens with all datafeeds I suppose.
From what I have seen when looking at the same datafeed it can vary slightly when you compare tick data (from which second charts are build) to the same feeds minute or daily data. For example the day high could be one or two ticks off - stuff like that. But this happens with all datafeeds I suppose.