How can you tell if an algo is at play?

[TimmyTP]

Someone made a comment today about algo trading being responsible for a 1 share bid on DIL.

I am sceptical about this, mainly because the 4,999 that were executed at that price is just too neat and tidy, but also because I don't think DIL is liquid enough to fit a volume curve reliably.

I'm not sure how I would use the information, but it might be interesting to explore a bit (we'll see if I'm alone in this). It can't be worse than calling out the latest price of your favourite/most hated stock all day.

So... does anyone think they can pick algo trading patterns reliably, using the data that is available to the general public? If so, how?

I'll try to kick off with some sort of contribution. For illustration - and since that's where the topic came up - I have used DIL (data is from 21st March 2014):
Attachment 5617

Target Volume
If someone was running a straightforward VWAP (same applies for some other common algos), the algo would likely have a target volume its active period. Commonly the active period would be the trading day. As such, I would expect a little burst of trading towards the end of the day, to meet the volume requirement and stand the best chance of meeting VWAP.

If I have attached my file correctly, you will see that there was no last-minute bust of trading activity.

Number of Trades
Another indicator I could think of is a large number of small trades, as the algo tries to meet VWAP without moving the market.
There was a total of 43 trades today for a total volume of 449,867, for an average volume of 8,331.


So by these two measures, my opinion is that there was no algo running on DIL today. That statement comes with a low degree of certainty, partly because I only have a single day's trading data to work with.

Anyhow, I'm interested in other opinions and advice. Also if anyone knows where the public can access historical trading data for NZX and ASX.

Finally - I searched around and couldn't find anything on this topic - I'll be grateful to be set straight if it's all been covered elsewhere, or if I'm on the wrong forum section.

Note: the figures I used here include the closing auction.

[TimmyTP]

Thanks for taking the time to reply Moosie. You made some interesting points - looks like not so many others share my interest in this topic, mind. We'll see I guess - this was definitely a tough one for Friday afternoon!

An algo can be as simple as the sp hitting a specified target and buying/selling everything up on offer

I think that example is a limit order, but I get your point -taken to the extreme, even stop, iceberg, IOC, FOK etc could be considered algos. I tend to associate 'algo' with more sophisticated techniques, since many exchanges provide the basic types natively, but it's true that anything a human does not explicitly initiate could be considered an algo. Where I see algo or 'bot' mentioned on this forum, my impression is that people are referring to the more complex types, that use past trading activity (and other, more sneaky methods) in an attempt to beat the market.

why would they need to meet targets at end of day?

I am thinking of orders from institutional investors. A common instruction would be "buy (or sell) me 500K TEL today at GVWAP"; institutional investors need to put there money to work on a schedule and failing to meet the volume target is a black mark for the broker*.
This in contrast to proprietary trading, where the bank holds the position on their own behalf and may have a more relaxed schedule.

I cannot see how any single individual could devote that much time to entering manual trades for one stock

43 trades in 7 hours must have originated from <=43 buy orders. Let's assume that all of those orders were entered by traders at NZX participants (imagine there's no DMA) on Friday; that's one order every 8 minutes, or one every 2 1/2 hours per participant firm. Even multiplying that across all 113 stocks on the NZX main board, that's one order every 1.3 minutes, across all the traders in NZ**.

is it that hard to believe it is a simple algo picking a long term price target for both the sell and buy side and trading accordingly?

it's not hard at all, but what I am struggling with is the confidence some people express when they state that an algo is active. Couldn't you equally well ask "is it that hard to believe it a retail order for 5,000 stocks, which happened to be hit by sell orders adding up to 4,999"? I want to learn what information people use to be so confident in distinguishing the two situations?

they definitely won that day in question with much higher volume on the good trades

This is a really interesting statement and exactly the sort of thing I'm trying to understand. What information or techniques did you use to determine that:
1. the buyer was running the algo, not the seller?
2. the buyer running the algo was the one buying the larger volumes?

So that's what I'm after - what do posters deduce (or what information do they have access to, which I am missing out on) to support such confident statements?

I was thinking about how this information could be useful to an average Joe like me. In certain circumstances, perhaps this knowledge could get you a better price. Let's imagine the spread is a big fat $1.50-$1.80 for something you really want to sell fairly urgently, but you know there has been an algo running that is likely to buy anything on offer above $1.62, but only from 13:30 to 15:00. With that knowledge, you could slap in a limit order at $1.62 rather than a market order which would execute at $1.50 or less...

* there might be a trader or ex-trader who could expand or correct this - I am neither, although I work in the industry and know quite a few
**these figures assume STLLH (Standard Trader Liquid Lunch Hour)

[arc]

Im Puzzled..

As a newbie in this game, can anyone tell me if the very first comments made in this thread in relation to purchase of a "single share" in a company is a "normal" thing..??
I have just done a search of day/week trade patterns via google finance and there appear to be "lots" of these happening... (1 to 6 shares)

1- Would'nt any purchase also have an associated "broker fee"?? ( $2 share with a $30 fee ?)
2- Can some-people themselves buy "fee-less"
3- Stupid conspiracy theory of backdoor/ insider software buying/collecting/obtaining below the radar (for free) and accumulating substantial holding till some future sell-buy date.


In relation to Algo, it would depend on the nature/objective of the program.

I found a site a while ago that triggered some thoughts (for me) on the Algo / Artificial Intelligence fields. http://rebellionresearch.com

Others::
http://en.wikipedia.org/wiki/Market_making
http://en.wikipedia.org/wiki/Algorithmic_trading
http://marketsmedia.com/series/machi...rading-series/


It all depends on how deep you want to go down the rabbit hole...


arc

[TimmyTP]

1- Would'nt any purchase also have an associated "broker fee"?? ( $2 share with a $30 fee ?)
2- Can some-people themselves buy "fee-less"

Typically you would get algos trading through two routes:
i. member trading
Orders are entered directly onto the exchange, by a member of the exchange.
In this scenario, a person or entity worth lots of $$ to the member firm would likely be offered a menu of algos to choose from, but all the client does is give the instruction to trade.
Alternatively, the firm might trade on their own behalf, holding the position for later resale, either onto the market or to their clients.
Generally members pay a fee based on the gross consideration of each transaction - see here for example.

ii. direct market access
In this case, clients of a member firm 'hire' the exchange membership and their systems for accessing the market. Generally this much faster than a retail platform (and is used with proper real-time depth information), so some clients can hook up their own algo systems to that. Clients of this status would be doing some fairly serious volumes and would usually negotiate a fee based on gross consideration - generally the more you trade, the less you pay per $.

3- Stupid conspiracy theory of backdoor/ insider software buying/collecting/obtaining below the radar (for free) and accumulating substantial holding till some future sell-buy date.

I'm not quite sure how this related, or what the question is. Certainly some algos can be used to hide intent to some extent.

In relation to Algo, it would depend on the nature/objective of the program.

I agree - in fact this is one of the reasons why I am so keen to understand how some people can be so confident that they have spotted an algo in action.

My point about the 1 share is that not all orders are executed in one go.
Imagine you have a sell of 5,000 shares and you're seventh in the depth behind sells adding up to 35,001.
Isn't it entirely possible that someone could buy a neat 40,000 (without the use of an algo engine), leaving you with one share?

Having said that, I am too dim to find the "day/week trade patterns via google finance". Could you post or PM me a link, please?

[TimmyTP]

Stumbled on this, thought it was quite relevant and interesting (to me at least!):
http://www.theatlantic.com/technolog...traders/60829/

[arc]

Tim

Ok, now I see the bigger picture. The "Exchange Members" effectively have internal access, and they can use Algos...

The article you linked is interesting. It sounds like the bots are generating deliberate high speed and also high volume "data packet noise" so as to disrupt "other" [competitors] computers as they deal with all the ficticious buy/sell orders. Distract/overload your competitors computers by a barrage of totally useless data to deal with, data that wont trade anyway as it is well outside the actual price range.
Its a smart move but also a dirty move... I wonder if they will make "laws" to stop it happening.


It sort of links into a research project im half way through, Creating a single large virtual computer out of many individual workstaions on a local network, Cluster computing, High performance with high throughput but at reduced costs [compared to owning a Mainframe].

Re Google Finance
The number of shares trading can be seen when you move your mouse over the graphs line. The data is displayed in the top lefthand side. The relative volume is displayed as bars below the day/date bar.

eg
https://www.google.com/finance?q=NZE...U6D-CoSmkgWdOQ

https://www.google.com/finance?q=ASX...U_jjJdChkgWuUQ

[TimmyTP]

Basically anyone buying and selling reasonable volumes of shares use alogorithms. So brokers, institutions, fund managers etc will all be running them. This enables them to buy/sell their shares over time without moving the price. Obviously you can't just dump a single order for 2 million shares in the market on one day and wait. So algorithms are always at play. They should not be confused with HFT which is just a special type of algorithm designed to do more than just execute a buy/sell order.

Individual investors can use them too if they trade using a good platform - eg. Interactive Brokers and Iress both offer algorithm functionality.

Additionally what you see on the ASX is only part of the market depth. An algorithm may be sending orders through dark pools with the instruction that anything left unfilled goes to the market. Normally what is on the market is only a small part of the order (known as an iceberg) as the order is being drip fed in tiny amounts over short/long periods.

Professional daytraders use algorithms as otherwise it would take too long to manually check your charts, check your price and depth, type out your order and submit it to market (one of the reasons novice day traders get slaughtered by the small minority of professional traders who do it for a living).

Brokerage is based on total volume not on each order, so makes no difference if you submit 1 order or 1,000 orders to market.

Thanks KW. I agree with all of that, except "algorithms are always at play" - because I was originally talking about individual instruments.
Some form of electronic trading is going on most of the time in the more liquid stocks, but that is not the case for less liquid stocks (so most of the NZX, for example). I was interested to know how some people were so confident in their observations that "bot" had started trading, but no-one has offered any basis for such bold statements, so I'm happy to make my own assumptions about that and diverge a bit.

So... on a slight side-track, it's interesting you mentioned Interactive Brokers. I have not looked into them before, but they offer a selection of standard algos that will run in Aus, including VWAP, TWAP, PoV and an API, so you could roll your own algos. Presumably they have DMA arrangements with various brokers, so you would be two steps away from the exchanges. They cover ASX and Chi-X Australia - but not little old NZX and it looks like they charge 8bps for ASX, if you are doing enough trading, plus some fixed market data fees up to about AUD70/mo.

Also I noticed your comment about SOR on the TRS/Mega thread. Again, (and as usual as far as I can see) what you say makes sense, although I wouldn't agree that ASX is the venue of last resort. It might not be the first (then again, I'm not sure it ever has been), but ASX now offers order types (e.g. Centre Point) that can achieve better execution outcomes, perhaps thanks to competition from Chi-X. If anyone is interested in a breakdown of how much is executed on which venues (including dark pools), please PM me.

*Already stated my "not strictly correct" view of what is an algo above.