How to discover profitable automation strategies?

[abgespaced]

I'm guessing with the amount of interest in automation that there are some strategies that actually work.

Not that I am expecting anyone to share their strategies. I am sure that the profitable ones are closely guarded.

What I would like to know is:

1. If there really is any opportunity to discover profitable automation strategies
2. How does one go about formulating a hypothesis to test

It would be great to hear from others if they are are making consistent long term profits with automation. That would solve point number one. However, I have my doubts. The markets are constantly evolving and no single strategy can remain static in the long term without losing its competitive edge to other people who end up running the same scripts.

Point number two- if there are indeed still opportunities out there, how do you go about discovering them? Is it just a byproduct of spending time trading in the markets? Do you develop a hypothesis and then test it using real money, considering that the only reliable results are real results? How do you even begin to ask the question of what to test?

Is this post even in the ballpark? Perhaps automation is simply an aid to manual trading and not a full replacement for it. The markets being efficient given a long enough time span, it just seems from my outsider's point of view that automation could either be wonderful, helpful or just a waste of time.

[wearthefoxhat]

I'm guessing with the amount of interest in automation that there are some strategies that actually work.

Not that I am expecting anyone to share their strategies. I am sure that the profitable ones are closely guarded.

What I would like to know is:

1. If there really is any opportunity to discover profitable automation strategies
2. How does one go about formulating a hypothesis to test

It would be great to hear from others if they are are making consistent long term profits with automation. That would solve point number one. However, I have my doubts. The markets are constantly evolving and no single strategy can remain static in the long term without losing its competitive edge to other people who end up running the same scripts.

Point number two- if there are indeed still opportunities out there, how do you go about discovering them? Is it just a byproduct of spending time trading in the markets? Do you develop a hypothesis and then test it using real money, considering that the only reliable results are real results? How do you even begin to ask the question of what to test?

Is this post even in the ballpark? Perhaps automation is simply an aid to manual trading and not a full replacement for it. The markets being efficient given a long enough time span, it just seems from my outsider's point of view that automation could either be wonderful, helpful or just a waste of time.

Always worth posting the question.

Having a hypothesis is probably the easy bit. Off the top of my head...
eg: Does the price of the favourite move in the opposition direction of the second favourite?

That's a fair starting point, it's logical and likely to be true and a reasonable supposition.

Finding the right market, timing for entry/exit, repeat or not repeat, automation/semi-automation? (those are the harder bits, but doable)

Here's a formula for success that can be applied for anything in life, trading would be no different.

TheFormula.jpg

learning.jpg

slowwalker.jpg

[Jukebox]

Here's a formula for success that can be applied for anything in life, trading would be no different.


TheFormula.jpg

I must update my Excel - I can't find any of those functions.

[jimibt]

Here's a formula for success that can be applied for anything in life, trading would be no different.


TheFormula.jpg

I must update my Excel - I can't find any of those functions.

yeah, I've had to do mine in .Net as I was similarly unable to locate.. (actually, jokes aside, this is actually a snippet from one of my strategies )

[jimibt]

OP, in terms of strategy research, I'd recommend that you think about a number of things:

1. Gather the raw data (either yourself or from the betfair data)
2. Record your sessions and review them with the raw data to hand
3. If you are able to, model what you are able to see into some form of coherent unit (i.e. simply break things down to the smallest moving piece, that may be something like -> identify contending runners. Maybe another one -> flag when contended order changes,, etc, etc
4. Commit to completing the lifecycle of an idea based on backtesting and live market participation
5. Identify where the process broke down and adjust using variants of the portion that is problematic

I know this isn't about identifying strategies per se, but is a fairly condensed view on how to discover if your ideas have got legs.

Others will hopefully chip in with more manual->automated type transitions. I personally always think from a mechanical perspective and thus seldom have any manual input on the ideas, other than maybe to exit something if it appears to be functioning incorrectly during a prototype phase.

[abgespaced]

Wow!

Awesome support, thanks guys. Just what I was looking for actually.

While I do have some experience coding and the automation side of things will eventually be important, I believe the quality in the process of testing ideas is what will define success in the end.

It certainly has a scientific feel to it. Having incomplete data or stopping short of an unequivocal result is counterproductive. You want to avoid that and build up a database of reliable results.

Thanks to all who have contributed to this post. It's gong to take some time to digest all this great info!

[wearthefoxhat]

Here's a formula for success that can be applied for anything in life, trading would be no different.


TheFormula.jpg

I must update my Excel - I can't find any of those functions.

The formula was taken from an old Supercalc3 file, so might explain it...or not...

[Trader Pat]

In answer to the title question... try, fail, try, fail, try, ooh thats interesting I'll make a note of that, try some more, fail some more etc etc... All the little bits you make a note of might eventually add up to a half decent bot. Definitely doesn't happen overnight

[jimibt]

... You want to avoid that and build up a database of reliable results.

tbh - more importantly, you want to build up a database (PRE and IP) of prices/stats for every race, every day so that you have total ownership of the data you personally rquire and in a format that is suitable for your modelling tools. as/when you explore new strategies, you should pump the data/parameters thro your historical races to ensure that both positive and negative outcomes tally to expectations.

this is a big undertaking at the outset, but once you have it up and runnning, it's mostly just Task Scheduler ploughing away, taking care of your active sessions and then archiving to the DB at the end of the day.

I actually find that setting up a generic graph that accepts data inputs (normalised to %'s) is a great way to spot little interactions that would normally fly stright over.

[Kai]

Fantastic opening post, asking the right questions for sure. Been thinking of opening a similar one to potentially discuss whether two traders can realistically use the same bots and about trust long term etc, but ultimately decided against it as I found all of the answers anyway.

1. If there really is any opportunity to discover profitable automation strategies

There are indeed auto opportunities on every market but some markets are more suitable to automation than others, duh.

2. How does one go about formulating a hypothesis to test

I'm sure everyone will say through data, that is how they find minor biases and patterns etc and take advantage of them for as long as they last, which may not be for too long because these can easily change as the markets evolve.

I don't really know anything about data and have zero interest in it, never used it for my trading anyway so I went with a completely different approach regarding auto, through collab with an active forumite on here that I've known for forever long before trading came along, and there are zero trading and personal secrets between us because we share absolutely everything. It made sense since he's a data guy and the ultimate quant, but we've basically skipped all of the data stuff since there is no need for any of that and we've simply been trying to directly utilize my order flow experience from day one, coupled with his Bet Angel experience because I wasn't a BA user, at first to try and auto some of my approaches but then through that process discovering new ones once I started looking at the markets through an auto perspective.

We've found Bet Angel a bit limiting when it comes to auto if I'm perfectly honest but it's still lightyears ahead of other stuff, ideally I assume you'd like your own software to do what you want to do, but we've still used Bet Angel to create a few different profitable bots, standard stuff for the most part plus a swing trading one which is my personal favorite due to my manual bias. They seem to be performing well from the start because they're based on existing edges so there's a lot of spillage, not as effective as manual or augmented trading but they can cover far more ground.

Point number two- if there are indeed still opportunities out there, how do you go about discovering them? Is it just a byproduct of spending time trading in the markets? Do you develop a hypothesis and then test it using real money, considering that the only reliable results are real results? How do you even begin to ask the question of what to test?

This would entirely depend on your market experience I guess. But I am personally not a fan of testing ideas that come from my head entirely, I'd rather get the ideas directly from the markets because they show you where the opportunities are when you actively study and trade them, that is where I get all of my ideas from, and sometimes from an occasional peek at another trader's ladder

After consulting with a couple of experienced botters it appears to me that genuine order flow experience seems to be key for them when it comes to auto, because apparently specialist knowledge is hard to come by for quant types with limited ladder time experience, they seem to look at everything through data.

That's just my 2 cents in a nutshell, I'm sure many will disagree with that but don't think I'll be changing my mind.

EDIT: Edited the oversharing bits, not on my own accord though

[jimibt]

... This would entirely depend on your market experience I guess. But I am personally not a fan of testing ideas that come from my head entirely, I'd rather get the ideas directly from the markets because they show you where the opportunities are when you actively study and trade them

yup - coming at it from the angle of the market via the data. this is where i tend to dwell. as mentioned previously, some crazy off kilter ideas will NEVER exist in your head until you either experience them in the context of the market - or if you review market related data in a new or novel way.

To come up with a FEEL for order flow via automation, you MAY for example choose to look at the back/lay side of the order book as a RATIO of one another, rather than hard numbers (i.e. orderRatioBest = best back money/best lay money [and ditto out to best 8 prices]) and compare the value change every n seconds etc, etc. This is a contrived example, but what I'm saying is that for automation sometimes ratios will work far more generically than hard numbers (equating everything to book% being a good example fo this also - can't sum apples and oranges ).

anyway, this is great (if endless) topic.

[Archery1969]

To the OP...

One thing you could look at is the following:

- Looking at the best 3 back and lay prices.
- Ratio of unmatched back money / unmatched lay money.
- Ratio of matched back money / matched lay money.
- Top, middle and bottom of a selections range.

If you combine the above then you will soon start to spot things and therefore be able to automate ideas etc.

Although I am probably giving a bit too much information away.

Have fun,

[wearthefoxhat]

Some good stuff on here for sure, albeit a little over my head on some of the more techy stuff.

Here's the performance of 1 strat I use, pre-covid (Jan-Mar), then post-covid (Jun-Aug). Profits are consistent, but the journey wasn't the same.
Seasonality seems to play it's part, but the market overall, tends to right itself over time.

Jan to March (jump racing and AWR)

beforecovid.jpg

Jun to August (AWR/Flat Turf)

aftercovid.jpg

[ShaunWhite]

Without staring an entire book on the subject the #1 sin with any data is backfitting. If you give me data I guarantee I can find you a strategy that would have won. What I can't guarantee is that it'll win ever again. Keep some data to one side for out of sample tests.

#2 would be an understanding of sample size with regard to statistical significance.

#3 Do as much testing and modeling as possible offline. Trialing or refining using money is a really ineffective way to do it and the final stage not the first one.

#4. Be realistic. If manual trading isn't easy then building something that will make you rich while you sit in the garden is even harder.

#5 Fail fast. Only 1 idea in (10?) will have legs so don't get wedded to ideas.

#6 Strategies that have a higher frequency will give you a faster yes/no than infrequent ones. If you trade 200 dogs a day you'll get am answer faster than say trading premiership footy with 11 games a week.

#7 for longevity and scale consider the fundamentals of any market and the things least likely to change. Things like supply and demand, or the tendency for markets to overshoot on price corrections amongst othees. That way it stands a chance of working everywhere dogs/horses/win/place/football or anything. 10p on 1000 trades a day is clearly better than 10p on 100 trades. That's also going to insulate you from the problem of killing all the fish. A strategy might start to fail on one market type but could still work elsewhere. Picking trap 6 on Wednesdays certainly won't do that.

#8 same again. If you're going to spend a year working on the perfect strategy, you want it to last otherwise the return never exceeds the investment.

#9 Expect a cost. For any idea you need to think of it as a raffle ticket. How much you spend depends on how much chance you think you have and the size of the prize on offer. (see #5)

That'll do for now.

Truth is though that there's as many types of automated trader as there are manual traders. Some automate what they've learned manually and focus on niche situations a bot can sit and look for (and you might not even need data for that) and some are hft quants placing 100s of 000s of offers a day for whom tick data and an accurate simulator are a necessity.

[ShaunWhite]

To the OP...

One thing you could look at is the following:

- Looking at the best 3 back and lay prices.
- Ratio of unmatched back money / unmatched lay money.
- Ratio of matched back money / matched lay money.
- Top, middle and bottom of a selections range.

If you combine the above then you will soon start to spot things and therefore be able to automate ideas etc.

Although I am probably giving a bit too much information away.

Have fun,

I don't think you've given anything away. There's only about 10 or 15 market metrics so naming 4 and saying 'combine them and you'll see something' is what everyone has been trying to do for donkey's years. If you can just 'see something" then that's a rare gift you have that not many people could hope to replicate.

Give us a clue archery, what weighing are you giving to each of those signals? what timeframe do you monitor? is the close formulaic too or neutral eg 0s or bsp? Do you flip your position if the signals flip? Offers or taking? The devil is in the detail.