Forums  > Trading  > gaming limit price  
     
Page 1 of 1
Display using:  

nikol


Total Posts: 751
Joined: Jun 2005
 
Posted: 2019-06-26 15:30
For discussion, I have a situation:

I want to buy at bid. tick size is 0.1. ask is at 8.8. Timeframe does not matter much.
1. observe 18 lots @ 7.8, next limit order has 4 lots @ 7.0 etc. LOB is tiny and sparse.
2. I set 1 lot @ 7.9
3. market immediately moves 18 to 8.0
4. I move 1 @ 8.1
5. market moves 18 to 8.2
6. I move 1 @ 8.3
7. market moves 18 lots to 7.8 (coming back)
Final state: my 1 lot @ 8.3 and market 18 lots @7.8

At this point, we both can repeat everything again from item number 1. (I did couple of times)
Given low volatility level, I can stay at every item quite long.

I imagine that:
- "market" algo estimates fair bid at about 8.2. Likely 8.3 is "too high" for him.
- by staying at 7.8 market prices-in additional margin.

Obviously, in the "Final State" if large sell comes, I will get through and my contender will be hit. Trades flow come in bursts, 1+1+1+20, or something like this (buy-buy-buy or sell-sell-sell).

Can you refer me to some related discussions/papers or explain in plain-vanilla how to deal with that?

rftx713


Total Posts: 106
Joined: May 2016
 
Posted: 2019-06-26 20:24
sort of unrelated but could you observe the behavior over time and track if this person is informed or not? sorry if that's a really stupid question

nikol


Total Posts: 751
Joined: Jun 2005
 
Posted: 2019-06-26 21:46
Informed in sense of "Garman"? I think, not, although I did not track this type of dynamics.

Additional info (realise it's important):
- I pay fee of 0.3% for market and get paid 0.1% for limit.
- In terms of speed we are more or less equally "slow".

rftx713


Total Posts: 106
Joined: May 2016
 
Posted: 2019-06-26 22:18
Not very familiar with Garman, started responding to your post when it still said Glosten Milgrom. (Any good links on Garman?)

I'm thinking in an illiquid market like this there likely aren't all that many participants (?). When you can spot a certain participant in the market and can somewhat game their algo to reveal its value estimate, it could be worth watching if it's an informed participant (or alternatively, uninformed) over a period of time.

Knowing somebody who has been right x% of the time has a higher/lower value estimate than you could be worth knowing.

I have no idea how to help on your original question, though. Apologies.

nikol


Total Posts: 751
Joined: Jun 2005
 
Posted: 2019-06-27 15:56
My apology for twisting the names. I was not sure.

Indeed, I meant this one (bad with names) Glosten and Milgrom (1985).
They translated a probability of Informed Trader into bid-ask.

I can almost definitely say, that my counterparty is not "informed trader", because he clearly reacts to my orders (in repeatable pattern) upto certain level X. When I cross this level and set my order at say X+e, he bounces back to the initial state. I don't see the rationale for this. My order is smaller than his, hence the next expected sell burst will trade through me for sure. He could stay almost without any harm.


deeds


Total Posts: 436
Joined: Dec 2008
 
Posted: 2019-06-27 17:13
Your very cool example and this general theme remind me of Larsen, Veiby and Timber Hill, of whom i am sure you are aware but mention just to be sure sure.

https://www.ft.com/content/e2f6d1cc-9447-11e1-bb47-00144feab49a


nikol


Total Posts: 751
Joined: Jun 2005
 
Posted: 2019-06-27 19:02
Applause

Yes, I remember it. Thank you for the reminder. Actually, I started to turn this problem in my head exactly in this direction - how to catch this bot.

My example is from a sweet unregulated crypto market.


rftx713


Total Posts: 106
Joined: May 2016
 
Posted: 2019-06-27 20:27
nikol - i think your potentially flawed assumption in saying he's not informed is your assumption that informed speculators have sophisticated execution algos. this could be somebody informed just using a stupid/flawed pegging algo or something like that. maybe?

edit: i just read your most recent post about it being an unregulated crypto exchange... and maybe am going to walk back my above statement a bit :)

ronin


Total Posts: 468
Joined: May 2006
 
Posted: 2019-07-05 12:59
> Can you refer me to some related discussions/papers or explain in plain-vanilla how to deal with that?

What is there to deal with?

If you are a liquidity taker, pull him in to improve his price and then hit him. It's nice that you get to find out what his best price is at zero cost to you. If it's not good enough, just sit back and wait until it gets better.

If you are a liquidity maker, you get his analytics for free (8.3 is loss) and you may or may not be more aggressive than him.

But he ain't very sophisticated. Sophisticated bots don't jump on one lot price imrovement.

"There is a SIX am?" -- Arthur

nikol


Total Posts: 751
Joined: Jun 2005
 
Posted: 2019-07-27 11:09
@ronin

Thank you for valuable comments (as usual).
After some investigation I see that my possibilities for profit are limited, except that I get analytics for free.

Now I wear shoes of that guy: it looks like I broadcast my analytics to the market myself.

My system trades in A and B and tries to catch the arbitrage between the two. A is more liquid with small spread, and B has the same liquidity but larger trades, less frequent and consecutively larger spread. I know, that my speed is at least as good as of others.

Time to time market jumps (5-10% move in 2 minutes), my system captures arbs (jumps is a source of ~40% of resulted PnL).

However, I face dilemma.
In order to catch the move I have to keep limit order on the market. That order price is a function of A + margin to cover possible fee and inefficiency (slippage) in execution. If I keep that order visible all the time I broadcast my analytics, which makes me nervous. If I enter only when arb opportunity arises, I am late and have to pay, so no arbitrage anyway.

What are the ways to obfuscate the strategy visible to the market?

ronin


Total Posts: 468
Joined: May 2006
 
Posted: 2019-07-29 13:38
I wonder if that is seriously a concern. There seems to be very little trading going on, so the cost of being read seems to be pretty much zero.

How do you hide your strategy - the usual. Randomize your size, maybe spread it around a few levels, stuff like that. All that has some cost in terms of per fill pnl, and you have to balance it against the cost of being picked off. But that seems to be zero in this particular case.

"There is a SIX am?" -- Arthur
Previous Thread :: Next Thread 
Page 1 of 1