How much is an ATS worth?

[kut2k2]

My question is, how does one go about objectively putting a price tag on an ATS?

Yes, I'm aware of the old adages that sellers should charge as much as they can get and buyers should pay as little as they can. Screw that. I'm looking for objective criteria to start with as a pricing basis. You don't walk into a new car dealership with $100 and expect to drive off with a new car. There is a reasonable range of prices.

But trading systems unfortunately seem to have no reasonable pricing method that I can find. The prices really are all over the place.

Is there an objective method for evaluating the true worth of an ATS? If so, where do we find it? If not, can we develop one in this thread?

I hope this thread will generate some positive feedback that is valuable to us all. Thank you.

 

[kut2k2]

To start things off, I propose we adopt the Surf Criterion. Marketsurfer is on record as stating that he would not trust the edge of any trading system based on fewer than 1000 trades. That sounds as good a place as any to start, unless somebody has a good reason to use a different minimum number of trades for generating performance stats.

So ... a minimum of 1000 trades .... for now.

Feedback welcome.

 

[Occam]

Interesting idea, but the answer for me depends heavily on whether you're asking what an ATS is worth sold as a stand-along product, or how much it's worth to a person using it themselves. If the latter, then I'd say it's worth exactly what the person can make from it, until it stops working, which is almost inevitable.

If you mean what it's worth in the "open market" to be sold as a stand-alone product, I'd say it's worth very little, if anything. Here's why:

There is almost unlimited capital available for an ATS that really works. With yields at 0, everyone is chasing alpha. Once the right people (e.g., at your broker, if you're using a decent one) see what your ATS is doing (in actual trading, not backtesting), they'll notice, and prop deals with vast amounts of capital will be at your disposal. So that begs the question: why is the ATS author not trading for his/her own capital or through a prop deal?

Also, I'd say that there is no "objective" value of an ATS based on backtesting (rather than a statistically significant number of real trades), because the ATS's value would depend heavily on who developed it, and how. You could have 1000000 trades in a backtest and still be selling a worthless, overfit piece of garbage. Not to mention the fact that there is often a huge disconnect between backtest results and reality, for a host of other reasons that might be called "slippage".

Then there is the question: if an ATS works, how long will its edge last? Short-term inefficiencies that are tradeable in an automated fashion are being wrung out of the system over time, especially since the mid-2000's or so, when electronic trading using open order books became the venue of choice for most executions.

 

[kut2k2]

Quote from Occam:

Interesting idea, but the answer for me depends heavily on whether you're asking what an ATS is worth sold as a stand-along product, or how much it's worth to a person using it themselves. If the latter, then I'd say it's worth exactly what the person can make from it, until it stops working, which is almost inevitable.

If you mean what it's worth in the "open market" to be sold as a stand-alone product, I'd say it's worth very little, if anything. Here's why:

There is almost unlimited capital available for an ATS that really works. With yields at 0, everyone is chasing alpha. Once the right people (e.g., at your broker, if you're using a decent one) see what your ATS is doing (in actual trading, not backtesting), they'll notice, and prop deals with vast amounts of capital will be at your disposal. So that begs the question: why is the ATS author not trading for his/her own capital or through a prop deal?

Hi, Occam

I have to disagree. It's possible for a trader to be cash poor and strategy rich. More to the point, just because an ATS works, that doesn't make it the Holy Grail. It may have an unacceptably large drawdown, or an unacceptably low winrate. What I'm trying to get at is a way of evaluating an ATS based on all of its performance factors, good or bad.

Quote from Occam:

Also, I'd say that there is no "objective" value of an ATS based on backtesting (rather than a statistically significant number of real trades), because the ATS's value would depend heavily on who developed it, and how. You could have 1000000 trades in a backtest and still be selling a worthless, overfit piece of garbage. Not to mention the fact that there is often a huge disconnect between backtest results and reality, for a host of other reasons that might be called "slippage".

Then there is the question: if an ATS works, how long will its edge last? Short-term inefficiencies that are tradeable in an automated fashion are being wrung out of the system over time, especially since the mid-2000's or so, when electronic trading using open order books became the venue of choice for most executions.

You misunderstand overfitting. It is impossible to overfit a system based on 1000000 data points, and very difficult to overfit a system based on 1000 data points.

And if the edge is time-limited, that can be part of the pricing procedure.

 

[kut2k2]

Let's look at the winrate.

If we express the winrate as a number between 0 (always loses) and 1 (always wins), I propose the value of the ATS should be proportional to winrate/(1 - winrate).

Feedback welcome.

 

[dtrader98]

Quote from kut2k2:

Let's look at the winrate.

If we express the winrate as a number between 0 (always loses) and 1 (always wins), I propose the value of the ATS should be proportional to winrate/(1 - winrate).

Feedback welcome.

The metric above is not necessarily good as you are only looking at probabilities of binary events. What would the turtle trader results have shown? One could also fairly easily overfit 1000 data points with a good grasp of optimization techniques.

 

[Occam]

Quote from kut2k2:

Hi, Occam

I have to disagree. It's possible for a trader to be cash poor and strategy rich. More to the point, just because an ATS works, that doesn't make it the Holy Grail. It may have an unacceptably large drawdown, or an unacceptably low winrate. What I'm trying to get at is a way of evaluating an ATS based on all of its performance factors, good or bad.

If you're "strategy rich", there's no need to stay "cash poor" for long. There's gobs of money available for trading a highly profitable automated system. "Downsides" such as "drawdowns are too large" or the "winrate unacceptably low" are only applicable if the expected return of the ATS is too low, in which case I'd call it worthless (or close), anyway.

You misunderstand overfitting. It is impossible to overfit a system based on 1000000 data points, and very difficult to overfit a system based on 1000 data points.

And if the edge is time-limited, that can be part of the pricing procedure.

It's possible to overfit any data set. Don't believe me? Send me 1000000 data points and I'll give you an algorithm that fits them perfectly. But I'll expect $1000000 compensation for this ATS -- it's got a lot of "value", right?

Re: edges fading, there's no way to accurately price it as it's far from predictable. So that brings me back to "trade it yourself" -- there's plenty of OPM if a strategy is really good. If it's not, maybe it's time to move on to something else...

 

[kut2k2]

Quote from dtrader98:

The metric above is not necessarily good as you are only looking at probabilities of binary events. What would the turtle trader results have shown? One could also fairly easily overfit 1000 data points with a good grasp of optimization techniques.

Winrate is not a binary event. I said a number between 0 and 1, not a number either 0 or 1. For example, if the winrate was 0.50, the factor would be 1; if the winrate was 0.75, the factor would be 3.

And wouldn't overfitting 1000 data points require using almost 1000 parameters? What kind of idiot would use a system that required fitting almost 1000 parameters?

 

[dtrader98]

Quote from kut2k2:

Winrate is not a binary event. I said a number between 0 and 1, not a number either 0 or 1. For example, if the winrate was 0.50, the factor would be 1; if the winrate was 0.75, the factor would be 3.

And wouldn't overfitting 1000 data points require using almost 1000 parameters? What kind of idiot would use a system that required fitting almost 1000 parameters?

You are right about win rate; I was thinking in terms of probabilities.

No, overfitting 1000 data points does not require 1000 parameters.

 

[kut2k2]

Quote from Occam:

If you're "strategy rich", there's no need to stay "cash poor" for long. There's gobs of money available for trading a highly profitable automated system. "Downsides" such as "drawdowns are too large" or the "winrate unacceptably low" are only applicable if the expected return of the ATS is too low, in which case I'd call it worthless (or close), anyway.



It's possible to overfit any data set. Don't believe me? Send me 1000000 data points and I'll give you an algorithm that fits them perfectly. But I'll expect $1000000 compensation for this ATS -- it's got a lot of "value", right?

Re: edges fading, there's no way to accurately price it as it's far from predictable. So that brings me back to "trade it yourself" -- there's plenty of OPM if a strategy is really good. If it's not, maybe it's time to move on to something else...

You're putting the cart before the horse. You're trying to declare a system as either worthwhile or worthless without doing an objective evaluation.

Maybe this thread isn't for you. This thread is for those thinking about either buying or selling an ATS. If that doesn't apply to you, thanks for dropping by and so long.