I have not looked at what perpetual contracts, you are dealing with. But if your rational is an impact cost of 1%. Maybe *your logic* because you are selling in drops of 1% of total position 100*1000K position 100.000K. After your one decade of trading experince! Or what! I guss that your perpetual contracts not are on Bitcoin, Etherum, Tether or Salanda. Because your guss of moving the market more than the daily vol many days, must be on some other more crypto exsotics. But, again with your own logic, why buy a 100.000K postion if you know you move the market up to the big nasty 100.000K sell order 1% abow the ASK. That with your logic *If I understood your right* will move the market more, than you drop buy and sell. Anyways way not jump over the topic of comision 200 trades are nothing to move 100.000K
Thanks, your ideas sound reasonable. I spent abround 10 hours going through articles like one you sent. At some point you just give up lol. Let's say, we have 2% edge for a given trade right now. We take a look at the TOB, ADV. We also know that average spread for the instrument and fees are 0.3% leaving us with 1.7% edge We want to leave some meat on the bone, so let's say we want to execute the amount that would account for impact cost 0.7% (both ways buy and sell), leaving us with 1% remaining edge after all costs. How do we figure out the amount we should execute based on that? Right now I use rolling 1% DOB as the amount. But is it even close to the truth? Since there are usually other 3-7 exchanges trading this perpetual future as well. And liquidity also is not distributed equally across order book: liquidity for nearest 0.5% from the spread is usually 2x-7x slimmer than same 0.5% liquidity but 2% away from spread, eg deeper in the book we go, fatter the book gets up until some point
Thank you for taking time to help me find solution. My post above describes situation in more details and answers your questions. You are right - trading goes for "exotic' altcoins top100-top500 of coinmarketcap
Congratulation with finding an edge. You second post is a bit more clear. I wrote about Jim Rickards a former lawyer and prem dealer * in one of my posts*, now writter. That loste some money in the bust of Long Team Capital Mangement. Jim claims that some crypto exhange create false liqvidity out of tin air. Is false liqvidity a real thing, that you know of, one any crypto exhange. I do not trade cryto myself, so I do not know! But, maybe something like CFD level 2 orderbook thing like a bucket shop
Thank you. Liquidity can be fake for sure, but edge was consistent for a very long time and my only worry is to estimate how much I can execute based on orderbook. I guess, the real question is give the edge and full information of orderbook - how to calculate relationship between amount executed and our impact. For now, I just take average 1% of orderbook depth in $ and assume that when I buy this amount in one taker order I will move market 0.5% up ( or my average price will be 0.5% up from best ask) And then , when later I sell it in one order I will also move it 0.5% down. So in total I will lose 0.5%+0.5% on impact. I understand how naive this model is since: 1. There are multiple exchanges trading same contract and they share liquidity. 2. Orderbook tends to be way slimmer around the current price 3. We can have fake liquidity 4. If we execute our amoint in many small tardes our impact could be less than 0.5% one way But I don't know how to incorporate all of that into my model.
Simply pushing the limits of size at one venue will generate the least overall profit. Plus your calculations are based on the whole order book which makes no sense, especially if stuffed to the rafters with ghost quotes. you need figure out the actual top of book. I'd try realtime T&S, make a model, get that right first. Then, you figure out max slice > repeat at other venues (not at exact same moment or size) > monitor impact > repeat at x interval until it reaches your threshold.
I understand your reasoning. How do I make a model that can approximate my impact? I use +-0.5% of last price to measure how much I can execute - is it too much? Let's say I bite 1-2 basis points per order - how would you measure your impact through the noise?
Price drift - on trade, pre and post (say 5-10 seconds) is the easiest. then monitoring spreads to return, order book to refill which if illiquid is prob more important. Yes 0.5% is a lot - you're gonna hoover up not just your impact but everyone elses so smaller. Maybe 5-10bps? Maximum size is only to gauge what's at level 1 btw - execution size should be smaller. You can learn a lot by watching how MMs do it in FX. The bigger trades take hours, sometimes days to fill so it ends up sliced into tens of thousands of pieces anchored to time or price, dynamic changes between time of slices and counter-flow to stop followers. Telegraphing by trading size is not common and usually either mistake/nefarious reasons.
I run a sim on bybit data I have since 2022. I have datapoints every 100 ms. At each timestamp we have info on best ask price and best bid price and also the size of contracts on these levels. Data is for 500 perpetual USDT contracts. Billions of datapoints. I tried to model your hypothesis. Let's say bid-ask spread is less than 0.0003 and we have some mid_price. In the next 100 ms best bid_price remains the same, but best ask_price explodes : goes up by more than 0.1%. Basically, someone bites a large taker order creating a spread - just how you assumed above. Then, we wait till the spread shrinks back to less than 0.0003, eg converges to new mid_price. I run this for many hours and got 1.5M situations like that. The idea was to see where the new mid price will move relative to old mid price and what % of that move will be made by ask price going down and what % by bid price moving up to fill the gap. As we see from pictures middle price always moves on average 50% of the gap created by ask price. Doesn't matter how drastic of a move ask price made the new mid price will go up by 50% of the move that ask price made. This works for any move of ask price from 6 basis points to 1%. Surprisingly, I don't observe any root law here. * Blue numbers on the chart are the sample size for each bin. I played with different instruments/ different dates - mid price always converges to 45-55% of the magnitude of ask price move, the relationship is very strong
Now it's getting interesting. What about over time? Seconds, minutes, hours later? Can you measure the same on other venues? It points towards a synthetic order book, but you need a bit more data to be conclusive. Or maybe just study that DOM - broadly symmetrical? Instant refills? Given they just withdrew from HK for 'operational reasons', if it *was* synthetic, they'd have had no choice.