Forums  > Basics  > Unintuitive thesis results  
     
Page 1 of 1
Display using:  

trialanderror


Total Posts: 49
Joined: Feb 2019
 
Posted: 2020-04-05 10:28
As some of you may have noticed I am writing my thesis, and I need to interpret a piece of results that is slightly unintuitive to me.

Suppose you submit an aggressive bid order to the exchange for less volume than what is offered at the current best ask at t. My results point towards there being quite a high probability of a market bid order following immediately afterwards at t + 1. I am not quite clear on what motivates this behaviour. One explanation I consider is competition between firms to take a good price first. Is this reasonable and are there other/better ones?

EDIT:
Another one.

Suppose a hidden order is hit on the bid side at t. The probability of a passive bid inside the quotes at t + 1 is also high. I'm at a complete loss here.

Nous promettons selon nos espérances, et nous tenons selon nos craintes.

gaj


Total Posts: 113
Joined: Apr 2018
 
Posted: 2020-04-05 10:50
Hate to quote Taleb, but this is narrative fallacy. Just because one thing happens after another doesn't mean the former *causes* the latter. Without the causation narrative, all you're seeing is two consecutive buy orders, nothing unusual about that.

Having said that, it is possible for the first order to tip the scales for the second buyer. This is just market impact.

trialanderror


Total Posts: 49
Joined: Feb 2019
 
Posted: 2020-04-05 11:01
Fair enough. However consider that I have split events into 20 categories. In the case of the aggressive bid with less volume than the best ask, the probability of there being a market bid order following is 15%, over 23 trading days looking at 30 index stocks. That feels suggestive of some type of relationship between the two to me. How would you word it to be more accurate?

Nous promettons selon nos espérances, et nous tenons selon nos craintes.

ronin


Total Posts: 601
Joined: May 2006
 
Posted: 2020-04-06 17:56
The first thing you describe is reaction to what looks like large bid volume coming up. Other bidders are rushing to get in before the avalanche.

The second one means that the best bid isn't where everybody thought it was - it is inside the spread. Bids are now improving to gain priority cheaply. You are the first in the queue, but you won't get hit until the hidden order is exhausted. That gives time for a nice queue to form behind you.

"There is a SIX am?" -- Arthur

trialanderror


Total Posts: 49
Joined: Feb 2019
 
Posted: 2020-04-10 16:45
@ronin thanks very informative

Nous promettons selon nos espérances, et nous tenons selon nos craintes.

EspressoLover


Total Posts: 449
Joined: Jan 2015
 
Posted: 2020-04-10 20:33
Scenario One: I go to the grocery store and see that the toilet paper shelves are nearly sold out. I decide to grab whatever's still left, even though toilet paper wasn't on my original shopping list.

Scenario Two: I'm interested in dating a girl at my office. We've been back and forth flirting for a few months now. One day she gets flowers from an anonymous admirer. I decide I better ask her out sooner rather than later.

Good questions outrank easy answers. -Paul Samuelson

nikol


Total Posts: 1195
Joined: Jun 2005
 
Posted: 2020-04-10 21:36
Second scenario is FOMO

trialanderror


Total Posts: 49
Joined: Feb 2019
 
Posted: 2020-04-11 10:29
@espresso nice to get some intuition for these somewhat abstract procedures :) now since this thread has gathered a bit of steam I'll throw another one out. It's a bit more involved.

I have split aggressive orders into multiple categories: walking (executing at different prices), large (price = opposite but more volume than offered at opposite) and medium (orders that behave equivalently to market orders).

Consider the bid side. After a walking or medium order, the probability of receiving a passive (bid) in or at the quotes is high. This makes sense to me, as I suppose that aggressive orders such as those are a sign of the mid being too low previously or something like that (that's what they said in the thesis I am replicating). However, when looking at large orders, the probability of getting a passive in the quotes is very low, at the quotes it's reasonably high but not very.

The only explanation I can come up with is as follows: at NASDAQ Sthlm when large orders execute, the unexecuted quantity is added as a passive at the order price. The "A" msg is counted as "part of the order" in my logic and so is not included in any of the passive in, at categories. So I assume that queue priority after a large order is automatically not good, and firms are therefore less likely to place order in or at the quotes.

A related curious observation is that for walking and large orders, it is very likely to observe a passive below the quotes, but not so for medium at all.

Nous promettons selon nos espérances, et nous tenons selon nos craintes.

ronin


Total Posts: 601
Joined: May 2006
 
Posted: 2020-04-11 19:38
> The only explanation I can come up with is as follows: at NASDAQ Sthlm when large orders execute, the unexecuted quantity is added as a passive at the order price. The "A" msg is counted as "part of the order" in my logic and so is not included in any of the passive in, at categories. So I assume that queue priority after a large order is automatically not good, and firms are therefore less likely to place order in or at the quotes.


Yeah, no.

The order type for aggressive orders is IOC. IOC stands for Immediate-Or-Cancel. Remaining liqidity is cancelled, it is not added to the book.

What you describe could be the effect of using a limit order to cross the spread. But that is not a serious tactic.

It is more likely to be somebody executing a vwap. He fell behind the vwap, so he had to cross the spread. Now he's caught up, so he can quote passively again.

"There is a SIX am?" -- Arthur

gaj


Total Posts: 113
Joined: Apr 2018
 
Posted: 2020-04-11 20:32
> What you describe could be the effect of using a limit order to cross the spread. But that is not a serious tactic.

If the spread one tick wide, I don't see the benefit of using IOC. If you send a limit order and have some unexecuted quantity, then either 1) somebody hit your remaining bid and the market goes back down, or 2) the market stays there and a queue forms behind your remaining bid. In case 1, you get filled for your full quantity. In case 2, you get a good queue priority. It's a win-win. IOC probably makes sense if the spread is several ticks wide, where an aggressive limit order would spook the market.


@trialanderror: Since you have the data, it sounds like you can easily check your own hypothesis, i.e., look at the A messages that come with the execution messages.

ronin


Total Posts: 601
Joined: May 2006
 
Posted: 2020-04-11 21:37
> In case 1, you get filled for your full quantity.

This reminds me of those sales pitches for carry. If the yen goes down, you make money in dollars. If the yen goes up, you make money in the yen.

Except, you don't. Both 1 and 2 lose money.

"There is a SIX am?" -- Arthur

trialanderror


Total Posts: 49
Joined: Feb 2019
 
Posted: 2020-04-11 21:39
@ronin I should have been more clear in my first post. What I call "aggressive orders" are just reconstructed from the ITCH feed, I do not observe the actual aggressive orders. I got some help from @EspressoLover with it here: https://www.nuclearphynance.com/Show%20Post.aspx?PostIDKey=196840.

Let me make an example of the reconstruction logic for large orders and medium orders, messages in () are within 1 ns of each other.

Large order (L):

ITCH message feed: [(EEA)AAAD(EEEA)DDUAE]
Reconstructed "aggressive order" feed: [LAAADLDDUAE]
Conditional prob to follow L order: 50% "A"-message, 50% "D"-message

Medium order (M): (assume that the "E" messages in the () reflect an "aggressive order" on the bid side, and that the combined volume of the "E"-messages = the ask side volume)

ITCH messages: [D(EE)DAADED(EE)UA(EEE)PA]
Reconstructed "aggressive order" feed: [DMDAADEDMUAMPA]
Conditional prob to follow M order: 1/3% "D"-message, 1/3% "U"-message, 1/3% "P"-message

@gaj Not sure how to check my hypothesis, I can simply observe that following the "L" event, there is a very low prob of an "A" message, which is strange considering the other results, and what prompted my gung-ho explanation suggested earlier.

I hope this makes sense.

Nous promettons selon nos espérances, et nous tenons selon nos craintes.

gaj


Total Posts: 113
Joined: Apr 2018
 
Posted: 2020-04-11 21:52
@ronin: Not sure I'm on the same page. Why would case 2 lose money? And if case 1 loses money, so does the IOC. Besides, an aggressive order is expected to lose money in the short term anyway.

ronin


Total Posts: 601
Joined: May 2006
 
Posted: 2020-04-12 09:11
> @ronin: Not sure I'm on the same page. Why would case 2 lose money? And if case 1 loses money, so does the IOC. Besides, an aggressive order is expected to lose money in the short term anyway.


The only reason you would be aggressing is because you are short - either in absolute terms (if you are market making), or relative to your benchmark (if you are running vwap).

In case 2, your remaning quantity on the limit order is bidding the market higher up than an IOC. That hurts your short.

In case 1, you aggressed too early and lost the spread.

There is no scenario in which this makes you money. Other than, you are really long and you are deliberately trying to bid the market up. And we all now how that one ends.

"There is a SIX am?" -- Arthur

gaj


Total Posts: 113
Joined: Apr 2018
 
Posted: 2020-04-12 10:10
if your IOC is not fully filled, how would you execute the remaining quantity?

EspressoLover


Total Posts: 449
Joined: Jan 2015
 
Posted: 2020-04-12 16:12
Maybe, I'm misunderstanding the conversation. But there are economic reasons to cross the spread, besides just inventory effects. Some participants have alphas, and sometimes those alphas both exceed the spread cost and decay too fast to be monetized passively. A good example is after the index futures tick up, it likely makes sense to sweep the touch on cash equities.

I also think there are also economic reasons to use IOC. Resting liquidity, even on new level formation, is subject to adverse selection costs in a way that pure IOC isn't. If I'm trading off a spurious alpha signal, others are more likely to quickly swipe my resting limit order. This goes double when you're not operating under latency supremacy, because you're probably arriving at Nth place instead of at the front of the queue.

Good questions outrank easy answers. -Paul Samuelson

gaj


Total Posts: 113
Joined: Apr 2018
 
Posted: 2020-04-12 16:42
Adverse selection on remaining liquidity is a valid point (and I think ronin might be alluding to a vaguely similar point). The argument is basically: if somebody hit my remaining bid in a newly formed level, he is probably informed. This may or may not be true for different markets and depends on how good my alpha is.

Some people use IOC just because of simplicity. They don't want to manage the lifetime of the order, handle weird order states, etc.

Kitno


Total Posts: 496
Joined: Mar 2005
 
Posted: 2020-04-12 17:45
@ronin yeh but gold goes up in both inflation and deflation! Applause

On a laager on a hill. A long way from Avondale.

ronin


Total Posts: 601
Joined: May 2006
 
Posted: 2020-04-12 19:57
> if your IOC is not fully filled, how would you execute the remaining quantity?

@gaj,
Say the market is 99/100. You swiped 100, the market is 99/101 and you are still not full.
Would you prefer to continue buying in a market that is 99/101, or in a market that is 100/101?

> But there are economic reasons to cross the spread, besides just inventory effects.

@el,
Yeah, I was probably being a bit too succint there. That was meant to be included in "short relative to the benchmark". In my world (when I was in that world), alpha saying go long creates a benchmark for how long you need to be and when. You are short relative to the benchmark, and you make your decisions on what's the best way to hit the benchmark in the given market. Decoupling alpha from execution was a thing then.

> @ronin yeh but gold goes up in both inflation and deflation!

@kitno,
Gold vs pork bellies is where it's at. I'll never forgive them for delisting pork bellies. Bastards.


"There is a SIX am?" -- Arthur

gaj


Total Posts: 113
Joined: Apr 2018
 
Posted: 2020-04-13 02:43
> Would you prefer to continue buying in a market that is 99/101, or in a market that is 100/101?

Fair point. The case I was thinking of was a liquid, thick market that's always one tick wide. 99/101 will either go to 99/100 or 100/101 in the next millisecond. If it goes back to 99/100, you should want to buy your remaining quantity again at 100, since nothing has changed from the last millisecond (unless you're an HFT guy, which goes back to EL's point). If it goes up to 100/101 you'd be happy to have good priority at 100, and you could always cancel it if you change your mind.

> Decoupling alpha from execution was a thing then.

All serious traders decouple alpha from execution. But in HFT world, alpha opportunities are usually instantaneous, so it doesn't make sense to execute with a schedule.

ronin


Total Posts: 601
Joined: May 2006
 
Posted: 2020-04-13 22:10
> Fair point. The case I was thinking of was a liquid, thick market that's always one tick wide. 99/101 will either go to 99/100 or 100/101 in the next millisecond. If it goes back to 99/100, you should want to buy your remaining quantity again at 100, since nothing has changed from the last millisecond (unless you're an HFT guy, which goes back to EL's point). If it goes up to 100/101 you'd be happy to have good priority at 100, and you could always cancel it if you change your mind.


Yeah, no.

You moved the mid in the adverse direction by half tick. You just paid half a tick to gain priority. Now think about how much this priority is worth.

In the good scenario, you get filled passively on both the bid and the ask. You made a total of one tick plus two rebates altogether, which is half tick plus one rebate on each side.

In the bad scenario, you get filled passively on your bid, then scratch. You made half a tick and a rebate on the fill, and you paid it back on the scratch.

So the game essentially has the expectation of quarter tick plus half rebate plus fudge factor. And you just paid half a tick to play it.


> since nothing has changed from the last millisecond (unless you're an HFT guy, which goes back to EL's point)

That is also a point where you want to pause for a moment. Actually, everything has changed. To quote ancient Greeks, you can't step in the same river twice.

You impacted the market, and the market will react in some way. How it reacts says something. Maybe somebody replenishes the 100 from the ask, which tells you something. Maybe somebody rushes in to open 100 from the bid. Maybe somebody crosses to 101. Maybe somebody crosses to 99. Each of those may mean something. And none of them tells you that priority at 100 is worth more than quarter tick.

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