Forums  > Trading  > better intraday execution  
     
Page 1 of 1
Display using:  

ahgt_123


Total Posts: 27
Joined: May 2020
 
Posted: 2022-02-12 17:55
I have developed a strategy with decent sharpe on intraday data (5-30 min) on an illiquid ticker. Upon live execution it performed poorly due to high slippage. (I target close price , i know its better to take mid price but i dont have that data).

How to develop better execution algorithms to reduce this slippage (I dont know where to start). I do have access to trades data for the said ticker but not quotes data. Any paper / book reference would be appreciated.

What have i tried.
1. quoting at best bid/ask
2. dividing total quantity into equal buckets and quoting at best bid/ask (this led to slight performance increase) for fixed time
.
Another legacy thread on this forum mentioned to seperate out the real signal and execution alpha. How would i proceed to do that?

nikol


Total Posts: 1474
Joined: Jun 2005
 
Posted: 2022-02-12 22:40
If you can then collect bid/ask/trade or even entire LOB from your feed yourself.
When you know the book, you can measure (survival) probability of your order execution as a function of price given the probability of your signal/trigger arrival. Find order price by maximizing PnL as function of that info.
Together with LOB/trade-info store results of your realtime measurements and actual executions to get an idea where you are.

By setting best bid/ask do you follow somebody else's quotes? So, if they move you move too as a dog tail?

PS. bottom line: without data I mentioned above you are in the dark.

... What is a man
If his chief good and market of his time
Be but to sleep and feed? (c)

ronin


Total Posts: 708
Joined: May 2006
 
Posted: 2022-02-14 10:43
I am not entirely sure I am following this.

5-30 min is your holding period? On an illiquid ticker? That does not sound very feasible in the real world. And then, what does the market close have to do with it?

Or, you are basing your signal on 5-30 min price data, and then getting out at close? Getting out at close is perfectly fine, if that is what you need. Unless your size is such that you are moving the close, in which case it is not. But I doubt you can really say anything based on that sort of data.

> 1. quoting at best bid/ask

And you are suffering from adverse fills. Your fill rate on winning trades is low, your fill rate on losing trades is 100%.

> 2. dividing total quantity into equal buckets and quoting at best bid/ask (this led to slight performance increase) for fixed time

Sound like half of TWAP. If you are not getting filled, you just go "forget it"? And you are still seeing the exact same adverse fills.

The good news is that you are on the right track.

The bad news is that you have a very long way to go.

A good place to start would be to search for key words like TWAP, VWAP etc, and work your way up from there.

"There is a SIX am?" -- Arthur

ahgt_123


Total Posts: 27
Joined: May 2020
 
Posted: 2022-02-15 09:01
Let me just clarify things a bit.

There is maker- taker fees in the ticker and the alpha diminshes if i use only market orders.

So, my problem is how to execute optimally with only limit orders while minimizing slippages.

I am familiar with TWAP and VWAP but aren't they generally used for end of day execution. I am looking for something higher frequency with large number of intraday churns.

@nikol i followed your backtesting advice and the only "signal" i have found is
1. execute as soon as you receive signal.
2. classify the last trade as buyer initiated or seller initiated and modify the quote a
bit.

I am still unable to cover the slippages incurred. Also, i have to add an additional "execution" latency which degrades the results further.

Any link to paper/material would be highly appreciated.

nikol


Total Posts: 1474
Joined: Jun 2005
 
Posted: 2022-02-15 10:34
If you switch your mind from waiting for the signal (e.g. plan vanilla "crossover"), to forward-looking estimation of how close you are to that, then you will get more info. Build sort of survival probability. So, 'receiving signal' is 100%. While approaching to that point you can estimate filling probability IF signal arrives based on your own view of the future at horizon-T. That implies MC, but maybe you can get it semi-analytically.

Filling is about knowing LO x MO interactions, isn't it?

On filling you are in market making domain. There are good threads here with literature. Software/Basic? Search within a year or two.

... What is a man
If his chief good and market of his time
Be but to sleep and feed? (c)
Previous Thread :: Next Thread 
Page 1 of 1