Forums  > Trading  > Trade impulse  
     
Page 1 of 1
Display using:  

liquidbread


Total Posts: 4
Joined: Jun 2022
 
Posted: 2022-06-28 00:17
Hi folks,

https://blog.headlandstech.com/2017/08/

I was reading the above blog post and under microstructure signal, there is this 'trade impulse' mentioned. I was hoping someone familiar with the formulation could clarify.

1) Where did the -0.25 come from? And why negative?
2) What is the denominator exactly? Post mentions the average of the bid and ask size but it is not clear to me what exactly that means.

Thanks

bbullero


Total Posts: 6
Joined: Jul 2021
 
Posted: 2022-06-29 17:58
Q1: Where did the -0.25 come from? And why negative?
A1: 0.25 is the tick size of the instrument in this example. Also, the next bid level is 25 cents below the best bid. So, if the current price level is exhausted the immediate price impact on the bid side will be 25 cents. In the example the average bid queue length is 15 units and 9 traded. Assuming the trade was informative, the trader who aggressed thinks that with probability 9/15 the price is too high and should be lower. With a stronger belief of the fair value the aggressor would have traded more.

Q2: What is the denominator exactly? Post mentions the average of the bid and ask size but it is not clear to me what exactly that means.
A2: This is just a data normalization technique.

liquidbread


Total Posts: 4
Joined: Jun 2022
 
Posted: 2022-06-30 05:41
Thanks bbullero for the response!
I suppose that "15" is the typical tob (1st level) size on the bid or ask?

Wonder if anyone has more insights on trade impulse used in application. As for the first point the blog post made, I have seen a paper by Stoikov.


bbullero


Total Posts: 6
Joined: Jul 2021
 
Posted: 2022-07-14 22:42
What do you need to know exactly?

liquidbread


Total Posts: 4
Joined: Jun 2022
 
Posted: 2022-07-19 15:06
Well, I am trying to learn/explore more about microstructure signals for trading spot fx. It can be either for market-making (eg better mid price construction/forecast) or liquidity-taking strategies, though I am more keen in the latter. Infrastructure wise, I don't have speed as an advantage (I am not at a HFT shop and I work on the sell side), so looking at alpha horizons of out to seconds and minutes. If you could be so kind, I am looking for some guidance and advice as to how do quant traders/researchers generally conduct research and set up the dataset for regression.

For a bit more context, presently, trying to replicate this (https://www.fixglobal.com/amp/short-term-alpha-signals/) with the dataset I have for a start. My dataset is a regular time series of 1second interval with the aforementioned quote imbalance and trade imbalance x-variables with y-variable being future 5min (or some other time in the future) midprice log return. I used rsq to evaluate my lasso regression (and could get a distribution of rsq depending on the iterations of using different time out into the future). I could do an actual backtest to see how it performs (ie open a trader every 1second and closing it 5min later to see how much pnl it makes).

Having structured my dataset this way, I think this might be ok to evaluate the viability of my x-variables using rsq(?), in reality, I don't think I want to be taking a trade every 1second and closing it out 5min later. Hence my question in the sense of how do you (or others in the industry) structure the research and dataset. So rather than sampling in clock time regularly, I suppose I sample in event time instead?

Pardon me if I sound convoluted. If it is, let me know and I will rephrase. I am just trying to give some background as to why I originally posted and what I am trying to achieve/learn.

By the way, I suppose you are still working in this industry?
Previous Thread :: Next Thread 
Page 1 of 1