Hi

I have a high frequency strategy that I'm testing. It utilizes bid ask data. Multi instrument strategy. It tries to seek arbitrage opportunities between two instruments. So for today's data. According to market replay in ninja trader i had 588 orders generated and executed. But in real time data (I have CQG) only 12 trades were generated and executed! My settings are identical for market replay and real time

Settings
Market replay = 1 min data series. COBC = false

Real time = 1 min data series. COBC = false

So in theory, since I believe market replay data is actually aggregated on 1 sec data, I was actually expecting more orders generated and executed in real time. But it actually generated significantly less trades!!! Why?

Could it be filtered data vs unfiltered data? Anyone know who the provider is for aggregating market replay data?

Any other reasons this could be happening??

Hi @calhawk01,

My guess is that Market Replay adds a lot of noise to a smooth market in real-time. I am not sure any high frequency strategy can be effectively testing on Replay due to the 1 sec buckets. We would need to know more about the trigger conditions to get you any closer, but my feeling is this might need to wait for NT8 with 1 millisecond resolution.

the strategy is simple. try to buy same instrument and profit from spread between different exchanges. pennies. according to market replay, this opportunity does exist (over 500 times). according to real time, it does not, well it does but on 13 times. i have been in contact with ninja support, they jumped on trying to provide support, but failed to understand what I am telling them, and the complain that they do not see any variance between market replay and real time. i'm going to work on creating an excel file that has data from market replay vs real time, and try to compare and contrast.

@calhawk01,

So does this spread difference exist for a few seconds or do the HFT kill it within 1 second?

seconds, maybe less than 1 second. Again, i can't see to verify if this difference exist in real time data. But I can in market replay data. That's the weird thing. So i'm thinking perhaps, the latency increases during real time, and perhaps the latency is probably close to zero on market replay data?

@calhawk01,

I would caution you that you may be tracking a digital artifact of time round off under Replay. You may not want to dig too deep into this issue until NT8 is our with 1 ms time stamp resolutions.

What could be happening during replay could be something like this:

INST1: at 10:15:45.305 @ $10, 10:15:45.805 @ $10, 10:15:46.002 @ $20, , 10:15:46.632 @ $20,
in Replay this becomes: 10:15:45 @ $10, 10:15:45 @ $10, 10:15:46 @ $20, 10:15:46 @ $20,


INST2: at 10:15:45.305 @ $10, 10:15:45.805 @ $10, 10:15:45.982 @ $20, 10:15:46.606 @ $20,
in Replay this becomes: 10:15:45 @ $10, 10:15:45 @ $10, 10:15:45 @ $20, 10:15:46 @ $20,

If INST1 & INST2 are correlated, real time arbitrage between the two prices around 10:15:45 is about 0.020Sec (20 ms), but under replay, this will show for a full second.