Practice Mode...in play realism?

Advanced automation available in Guardian - Chat with others and share files here.
Post Reply
Stubbo
Posts: 56
Joined: Fri Dec 02, 2016 8:44 am

Have come up with an automation script that works very well in practice mode during In Play racing events (using 10 GBP stakes, so nothing that to my knowledge would significantly impact a live market which is obviously a key consideration moving from practice to live).

However, when going 'live' with it, an entirely different experience ensues.

From what I can see this is largely to do with the bot in practice mode managing to Green Up very successfully, but in live this is rarely the case...also if I run practice and live concurrently, bets are not placed at exactly the same time using the same automation file running on the same machine with the same connection speed.

From others experiences would you say that this is fairly typical? Having strived to find something that 'works' this is obviously rather frustrating.
Bluesky
Posts: 420
Joined: Mon Sep 19, 2016 9:26 pm

I'm sure one of the experts here will give you a really good explanation. The support on this site is terrific and is the main reason I chose Betangel over the several other similar products available.

One thing I would probably try if I was you was to use the minimum stakes (£2) and see if that will bring your results closer to the practise mode ones. There are many strategies that just wont scale up, I know £10 stakes aren't exactly huge, but it could be the case that perhaps your system works fine with £2 but not with £10.
User avatar
LeTiss
Posts: 5386
Joined: Fri May 08, 2009 6:04 pm

Practice mode isn't my cup of tea at all

It doesn't accurately reflect markets at all with regards to getting fully matched, and it also doesn't indicate how your mindset will change when playing with real money.
Stubbo
Posts: 56
Joined: Fri Dec 02, 2016 8:44 am

Thanks for the reply...I actually tried this and had the same experience with £2.00 stakes too.
shakespeare
Posts: 113
Joined: Sat May 21, 2016 1:43 pm

It may just be that your practice mode experience was not long enough for you to see the problems you're seeing in live mode. I've never noticed any significant difference between practice mode and live mode. I lose the same amount in both. :lol:

It's perfectly normal for two identical bots to post different results. If you ran the same bot in two practice mode sessions at the same time, you would not get the same results, because the markets move so fast. In a way that shows you how good practice mode is in simulating reality.
User avatar
Euler
Posts: 24702
Joined: Wed Nov 10, 2010 1:39 pm
Location: Bet Angel HQ

Practice mode is similar to taking a penalty against a blindfolded goalkeeper. It's an accurate representation of most things, but not how the market reacts to a position. The goalie can't see which what you are shooting or guess at it, but the market can.

It's nearly always best to use minimum stakes to test a strategy, even if it's to work out the fill rate on certain strategies.
Stubbo
Posts: 56
Joined: Fri Dec 02, 2016 8:44 am

I get what you're saying Peter, but equally would suspect a £2.00 bet at minimum stakes in a market where over £1,000,000 gets traded would be pretty much under the radar (especially if focused towards a well traded horse within that market), and whilst I understand the blindfolded goalie analogy, struggle to believe that the market reacts in any way to a £2.00 bet placement.

Leads me to believe the issue is more likely to be with respect speed of placement in the market (does practice mode build in the 1 second delay to market placement?) or a question of latency (e.g. a bet in practice mode doesn't need to 'go' anywhere...it resides on the machine, versus a bet placed in live needs get from the Guardian decision making process into API form, to travel from my house to Betfair servers and get processed which likely increases the 1 second delay once received by Betfair even further).

It doesn't really explain though why attempted bets are not identical.

Given practice and live would be receiving the same stream of data and so have the same inputs, would trigger at the same point in time, and have the same logic in place to make decisions, you'd expect the same decisions would be reached unless either:

- they are receiving slightly differentlt timed data streams
- the computer's processing capability can't handle the volume of data/decision making and was struggling to keep up
- the frequency with which Guardian code can check the market data and make the logic decision to open a bet is variable at the 20ms level

Anyway it's all moot as clearly they aren't the same and don't behave in the same way even at minimum stakes whatever the reason, which is a learning in it's own right.

Also it make me wonder if decreasing the stream rate would yield more consistent results...given the latency and API performance will by default be variable, the built in 1 second delay, and other areas of possible inconsistency, making sub-second decisions doesn't neccessarily seem to make a great deal of sense as even if you can spot opportunities at that rate they can't be taken advantage of (as a lot of price movement can happen in a second) so smoothing the data out and basing strategies on greater data refresh increments (certainly >200ms) seems logical.

Thanks to those who replied above for your insights.
xitian
Posts: 457
Joined: Fri Jul 08, 2011 2:08 pm

Unfortunately what you've experienced is quite common - at least in my own experience too.

Spend a lot of time researching, backtesting, simulating a strategy, and it all looks good. Then run live with real money, and you'll find it breaks down. Try not to be too frustrated. I just suggest in future not to get too optimistic until you've got to the stage where it's actually making real money. At some point, if you're persistent enough, you'll find an edge which does work in live markets. Also, you may find that quite minor modifications can make significant differences. Execution is probably an overlooked (or should I say underlooked?) area of developing trading strategies.

With regards to your practice mode question, I can only answer from the standpoint of someone who writes their own API code rather than using BetAngel automation, but I've found that it can be very hard to accurately "simulate" placing/matching bets particularly for latency dependent strategies, and particularly for in-play horse racing. I would say that the main source of simulation error would come from not modelling the latency of bet placement. Even a few ms here and there will make the difference between a successfully matched bet and one left unmatched (and if unmatched, where you are in the subsequent queue). However it's pretty impossible to model accurately as you have no idea what's happening in that 150ms or so that it takes Betfair servers to return the status of your bet. At the moment in my backtests I assume a fixed latency in my simulations, but I can of course test my strategies at difference latencies to see how "sensitive" my strategies are to it. It would be a cool feature for BetAngel if you could customise that latency for practice mode.

I agree that there's a bit of curiosity surrounding why it's not firing the same bets though. All I can suggest is that you get clued up on programming yourself and start writing your own code. You have far more power that way, and can determine/inspect instruction by instruction what your programs are doing.
LinusP
Posts: 1871
Joined: Mon Jul 02, 2012 10:45 pm

xitian, I encounter exactly the same issue when it comes to Inplay racing. It's so volatile that ms really do matter when it comes to getting matched, hadn't thought about adjusting latency to check for sensitivities so thanks for that tip.

I am fairly sure betangel does include the bet delay in practise mode but not sure if latency is included. From what I remember it is 'under sensitive' to being matched but will sometimes match the full amount if the price trades past your order which can be unrealistic.
mtrehearn
Posts: 52
Joined: Thu Sep 23, 2021 9:51 pm

LinusP wrote:
Mon Feb 20, 2017 3:01 pm
xitian, I encounter exactly the same issue when it comes to Inplay racing. It's so volatile that ms really do matter when it comes to getting matched, hadn't thought about adjusting latency to check for sensitivities so thanks for that tip.

I am fairly sure betangel does include the bet delay in practise mode but not sure if latency is included. From what I remember it is 'under sensitive' to being matched but will sometimes match the full amount if the price trades past your order which can be unrealistic.
I am running in to the same issues with my in running lay bot. Different horses are sometimes selected meaning I have losses in the real world that otherwise would have been wins in practice.

Did you ever get this fixed?
Last edited by mtrehearn on Tue Jul 05, 2022 10:08 am, edited 1 time in total.
mtrehearn
Posts: 52
Joined: Thu Sep 23, 2021 9:51 pm

Stubbo wrote:
Sat Feb 18, 2017 7:24 am
Have come up with an automation script that works very well in practice mode during In Play racing events (using 10 GBP stakes, so nothing that to my knowledge would significantly impact a live market which is obviously a key consideration moving from practice to live).

However, when going 'live' with it, an entirely different experience ensues.

From what I can see this is largely to do with the bot in practice mode managing to Green Up very successfully, but in live this is rarely the case...also if I run practice and live concurrently, bets are not placed at exactly the same time using the same automation file running on the same machine with the same connection speed.

From others experiences would you say that this is fairly typical? Having strived to find something that 'works' this is obviously rather frustrating.
I am running in to the same issues with my in running lay bot. Different horses are sometimes selected meaning I have losses in the real world that otherwise would have been wins in practice.

Did you ever get this fixed?
User avatar
ShaunWhite
Posts: 9731
Joined: Sat Sep 03, 2016 3:42 am

mtrehearn wrote:
Tue Jul 05, 2022 10:07 am
Did you ever get this fixed?
😁 Those posts are 5yrs old..... And Liam, xitian (and myself) aren't using BA but still talking about the same considerations so it's not a BA problem it's just the nature of simulations. There is no "fix'
Post Reply

Return to “Bet Angel - Automation”