Forums  > Trading  > orders distribution vs forward price and trade distribution  
     
Page 2 of 2Goto to page: 1, [2] Prev
Display using:  

nikol


Total Posts: 576
Joined: Jun 2005
 
Posted: 2018-09-26 11:01
Understand.
Thank you for your advises.

nikol


Total Posts: 576
Joined: Jun 2005
 
Posted: 2018-10-01 19:03
Order book monitor

python

pyplot.scatter with markersize = order_size
pyplot.plot red line connecting best bid and best ask

not yet fully interactive, but very informative




ronin


Total Posts: 385
Joined: May 2006
 
Posted: 2018-10-02 10:33
What's the timescale on this one?

There is an obvious seller and an obvious buyer, both working their orders. The seller is more aggressive. They don't seem to be hitting each other - they are both improving until somebody random hits them. They also both seem to be blind - neither of them is reacting to the other one.

The orderbook is pretty sparse and pretty static. So what is the typical order size - is it one clip per level? Try colour coding for sizes, this doesn't really show much.

It doesn't really scream liquidity. So I wouldn't really expect you to be profitable with any sort of high to medium frequency approach.

"There is a SIX am?" -- Arthur

nikol


Total Posts: 576
Joined: Jun 2005
 
Posted: 2018-10-02 19:24
Remember that hitting (taking) is cost. making is not.
Time frame is about 10 second

Price slot volume:
Smallest size (at best line) is about 1 EUR :)
And the biggest one at this market (ETH-EUR) within this price window is at 50k EUR.
At BTC-EUR you can observe as big as 25k BTC's which is at current prices is around 150 Mio EUR.

Seller and buyer can be same trader.
In other time frames (2 hours later) I see that both are moving their smallish orders of ~1 EUR size in sync leaving very small spread in between, while often there is large empty space behind bid and ask with larger orders of few 10k EUR size.

Colouring idea is interesting, it seems to be possible. Now busy with some other urgent project, will come later.

Look at the picture above. I have updated it.

ronin


Total Posts: 385
Joined: May 2006
 
Posted: 2018-10-04 14:28
> Time frame is about 10 second

So these guys are going "OMG I didn't get hit in 100 miliseconds must improve OMG I didn't get hit in another 100 miliseconds must improve OMG I didn't get hit in another 100 miliseconds must improve" etc, while the market is doing nothing?

This is weird af.

The fact that there is something that looks like liquidity in deep book is more or less meaningless. It only matters if it's still there when you get to it.


"There is a SIX am?" -- Arthur

nikol


Total Posts: 576
Joined: Jun 2005
 
Posted: 2018-10-04 16:06
I understand your irony, but that's the slow life of this exchange... maximum you can have is 50 messages per second = in average 20 milliseconds under control.
Our setup allows 0.1 ms bouncing. If we stretch, perhaps we can go below 0.05 ms, but that's the major exercise.

Here is another time window of ~10 s.

Vertical sets of dots represent bits of information, which is tick update somewhere in the book not necessarily within this price range.

Such sawtooth pattern repeats 5-10 times regularly here and there at different time-scales and amplitude. However, amplitude does not exceed fee amount ;)
Already thinking about your earlier suggestions to scalp him.



This bar-plot is slice of log_volume(price) at time = 50 above.
Highest point here is ~226 ETH ~ 47k EUR
Best (front line) orders are ~1 ETH ~ 210 EUR




ronin


Total Posts: 385
Joined: May 2006
 
Posted: 2018-10-05 14:48
I'm not being ironic, I am genuinely trying to understand what is going on.

Your graph below is basically showing that you would pay 30 bps plus fees to roundtrip say 5k euro. 20k euro, who knows - it's outside the graph.

And you can't scalp the guy because he is only throwing 20 euro or so inside the spread. You could be scalping him all day and it wouldn't pay for a pint of beer in the evening.

So I would say this high frequency approach is fruitless at this stage. There is nothing you can make on this scale. Look at longer time scales, maybe something comes out.


"There is a SIX am?" -- Arthur

nikol


Total Posts: 576
Joined: Jun 2005
 
Posted: 2018-10-05 15:15
Ok. Then I will take a look to other exchanges.. There are about 20 of similar liquidity and wider spread.

nikol


Total Posts: 576
Joined: Jun 2005
 
Posted: 2018-10-19 11:50
These are execution times:

https://news.bitcoin.com/order-speed-analysis-reveals-the-fastest-cryptocurrency-exchanges/

To my view execution delay is not that big problem in such delayed market since I can safely predict when order will arrive.
Huge variance of that number is main problem.

And I am wondering how other guys are able to draw such beautiful "saw" patterns

Cry

nikol


Total Posts: 576
Joined: Jun 2005
 
Posted: 2018-10-20 20:50
Can it be that the single matching engine is running on several servers?

ronin


Total Posts: 385
Joined: May 2006
 
Posted: 2018-10-22 11:42
> Can it be that the single matching engine is running on several servers?

I don't really know about crypto echanges.

In regular exchanges, there is a single tape. Orders are processed in the order in which they arrive. Of course there are multiple servers doing the processing, but it's a single tape.


"There is a SIX am?" -- Arthur

bullero


Total Posts: 26
Joined: Feb 2018
 
Posted: 2018-10-22 13:11
nikol, what is this "saw" pattern you are referring to exactly? If you refer to the triangles or "saw" like price action in the pictures you have posted previously, my first guess is that those are created by a simple execution/market making algorithm. To be more precise, those remind me of HJB style optimal bid and ask prices with given time constraint. Sometimes the pattern emerges at higher frequency which would be indicative of the buyer being in a hurry.

nikol


Total Posts: 576
Joined: Jun 2005
 
Posted: 2018-10-22 13:47
@ronin

Single tape concept is clear. And I think it plays the role of bottle neck for the entire system, because locally one have to make sure that the order is maintained properly across the entire book. Perhaps (I dont know exactly), queues within price slots are run independently, while single match/cross-engine is taking limit orders and writes trades on to the tape.

Major crypto-exchanges are the same. But there are some with different concepts, like Bitfinex. On this one people trade crypto on margin, but this margin demand they have to market on another 'funding exchange' also run by Bitfinex.
Some decentralized exchanges are coming too.

Meanwhile, we moved closer to the exchange (it took sometime to find where it is) so execution times improved significantly (about 10 to 15 times). Now we will collect data from this new place and I will have to repeat analysis in search for arbitrage and trade signals. At least, I am now confident to be able to perform the same "jumpy" quoting.

nikol


Total Posts: 576
Joined: Jun 2005
 
Posted: 2018-10-22 14:18
@bullero

Yes to all your guesses.

"pattern emerges at higher frequency which would be indicative of the buyer being in a hurry." - not completely true. I would use such pattern to 'obfuscate' market signals for other market participants.


Thank you for this "HJB style optimal bid and ask prices with given time constraint". Very interesting.

bullero


Total Posts: 26
Joined: Feb 2018
 
Posted: 2018-11-16 21:18
Any new ideas / discoveries here?

Oh and btw, could you elaborate what would be the motive to obsfucate the other participants? You refer to "leaning against" passive liquidity placed by other guys? For example, in the blue saw-pattern the price climbs fast up after being smashed and repeats several times in sequence. One thing that could be going on there is that lets say you want to sell but dont like the price you place tiny order by improving price and if the other guys algo is designed to be at best level he will improve and you repeat until price is such that you dont take too big a hit when hitting his bids.

nikol


Total Posts: 576
Joined: Jun 2005
 
Posted: 2018-11-17 00:41
Why obfuscate?
If your opponent is AI (or cumulative I) is trained to the best asks(bids) fluctuations, it will naturally stay away from the best front.
Is it reasonable enough?

Or. Obfuscation of order_book information as indicator of future price movement has also certain benefits.

"One thing that could be going on there is that lets say you want to sell but dont like the price"

As you can see from plots, nobody cares about jumps of these quotes. Entire picture is still. I did zoom into trades, they are also not much related.


No discoveries. Data are understood. I deviated a bit from this, by digging into different direction (analysis of daily data). Meanwhile, I got idea on how to clean these oscillations off, so about to test it.

Previous Thread :: Next Thread 
Page 2 of 2Goto to page: 1, [2] Prev