This is my first question in here, so please excuse any eventual mistakes (wrong forum, etc)...
Anybody has or know where to find reliable piece of code that would:
1) construct good time-based bars from ticks?
2) construct bigger timeframe bars from 1 second bars?
I already have a piece of code that does that, but I would really love to find something more "official", or already tested...
What would be the reliability of such a conversion?
Would it be better to save the already formed bars provided from the datafeed (although not very practical because there is virtually an infinite number of bars (ranges, time-based, ...)
If you have tick data and store the tick data (provided you have the space) then you can create any kind of
time based bars (and all other synthetic and derived bars). Storing tick data gives you the broadest option.
1. Is your data good tick data ? Do you need maximum flexibility ?
2. Alternatively you can decide to aggregate tick data into an acceptable time frame bar, the fundamental
difference between tick and time based, is that you will have to let go some other information, which you
might have or not...
On a tick level, you might have bid and ask price and you might have the volume at the bid and ask, if
applicable.. That information is useful in some cases, to create certain extra insight...
1. How can I check if I have good tick data? I am starting from AMP's 5-year history offered along with Multicharts special edition. Is there any better affordable solution? I only need 2 years of tick history.
2. Pre-agregating it in bars doesn't feel right...
I am building a new platform and I need at this point to build a historical data too, otherwise it would be very counter-producting to write for MC or NT and backtest there and *then* migrate code to my platform. The backtester (trades engine) is already written, auto-trading part too. All I need is historical data.
My idea is to put all the ticks (also the best ones ) in Google BigQuery or the Google Datastore and then requesting them when backtesting. Everything would run from Amazon instances so internet speed would never be an issue.
My strategies trade futures intraday and work with 2 timeseries of the same future working all like this:
Let's imagine it goes long on green inside-bars and short on red inside-bars. It would wait for a 15 mins inside bar to close then send stop-limit order with the next tick (or the close of the first 1 sec bar after the close of the 15 min inside bar).
Every strategy needs a big timeframe (like 15m, 5m, 60m for geometrical recognition purposes) and a "clock" one (tick, 1 second or at worst 5 seconds). I also trade up to 6 different symbols.
You can immagine that given this situation my historical database must provide me any kind of bars for the geometrical dataseries and at worst 5 second bars for the "clock" dataseries.
This actually makes my day! So simple and I didn't think of it
I'll store Ticks and 1 Second bars and maybe, just maybe 1 min bars and daily bars as provided by the source, so that I will be able to build any type of timeframe without having to actually request an infinite amount of data (it would be a sin to load ticks in order to build 1 year worth of daily bars...)
I will use the ticks to build all kind of bars, but time-based ones.
... the 1 Second bars to build up to 1 minute
... the 1 minute bars to build bars up to 23 hours
... the daily ones up to 1 year (will have only 2 years though haha)
Lastly, anybody knows anything about the quality of historical data provided by AMP with Multicharts Special Edition through CQG? How would you measure it programmatically?
Is IQFeed better (saw it mentioned around here a lot)?