Australia's largest stock trading and investment forum Australia's #1 stock forum

XJO

Click for more information

0.88%

7,959.3 s&p/asx 200

Indices 2/10, page-282

First 282 Last

Share
Profile

Follow
Launched

2,398 Posts.

2603

07/10/23

16:01

Post #: 70225271
Share

Originally posted by Coincacha: ↑

Launched there a lot to unpack here, so I'll keep on point as best I can.

There are many ways to use and benefit from automated systems that don't all equate to be being wildly profitable by themselves. It's a full on journey but the the other side of it is pure bliss.

I use PRT(IG) and it includes in house code and training courses. It took me 6 months of intensive work to get a handle on coding as it was not my background. So, whichever way you go be it Python, c++ or in house languages you need to know where you are trying to end up. I live with chronic pain and home school 3 kids so Auto was largely a path that was necessary. The biggest advantage I have learnt to learning a real language like Python would be the flexibility of incorporating multi instruments like volatility into advanced algorithms that use Dynamic Position Sizing (DPS). As I use Probuild I have had to build an indicator I call ATR Zones which takes percentage atr and overlays a Fibonacci sequence to tell me when I should be looking for Mean Reversion, trend or on-trend swings and how hard to hit it giving me an idea of expected return. This has almost been the single biggest benefit I have experienced so far as had allowed full conviction on large size trades I would not have otherwise had.

Chat GPT can certainly be useful for efficiency to someone who is already highly competent in the language of choice but will produce errors that will likely not be picked up until live testing. I have observed this on many occasions and do not use it myself, just tested it as a matter of course. That said I'm sure it will get far better very quickly.

Back testing is a huge expertise in itself and riddled with pitfalls you need to be privy of. As with portfolio manangement. I have had most success with hourly entries on a Daily time frame (slow bots/trend) and 1 min entries for higher frequency pattern exploitation (maybe trading every 1-3 days). These types of bots use price rather than indicators and become very mathematical as they rely on velocity and acceleration. You never want/or should build the algorithm on the full data set. It needs to be valid on less than half available data then tested using OOS, or out-of-sample data. The lower the TF the less data you have to work with which is very dangerous. But I've found no other TF to be productive. I want production. You can save yourself a lot of time by reading the many articles available on this topic. Not doing this is the definition of curve fitting! If a bot needs to be tuned regularly it is curve fit. Can still be used if YOUR the main filter and you are using it to alert you of trades but you take it manually.

Other truths you need to be aware of are the statistical probability of cluster wins and losses are far more common that you may like (really testing your conviction in your work). The only way around this I have found is to not run too many bots at once as it's becomes a nightmare to monitor and review the systems as a portfolio. In this case Win Rate does matter, but only for this purpose and testing entries.

You've said previously that you would like to pursue something that can add to your bottom line and take a bit of workload off. This is the way to go if you have full availability to trade manually as your main business income. However it is true unfortunately that it'll take about 50 to 100 bots to find a really good one that is robust and you can trust.

As it stands now, I run a long/short pair both independently profitable and negatively correlated on the NDQ as my high stakes account. This is for outsized growth and represents 40% of my portfolio. I don't have to touch this account. It has passed Monte Carlo testing and OOS to flirt with Risk of Ruin divided by 2. So it can't blow up during cluster events and has a return to DD ratio of 26 as I use this as my main metric of profitability and the journey to get there. Most surprisingly it is a very low WR system ranging from 28-45% but profit factor is through the roof.

My second account is full of B-grade bots that automatically switch on/off inline with market conditions. These produce reliable income on a quarterly basis but each only trades a couple of times a month. I live test here with single contracts and the proven systems are risk adjusted and the portfolio finely balanced through correlation studies. Largely time based bots to maximise the margin availability of the account. I'm allowed to intervene with 3 of them for profit taking to get best price manually and I log this to make sure which systems I can trade better than. Again, 40% here.

The last account is for manual. After now 18mths of doing this and taking the time off manual trading the question remained. Have I become a better trader through studying probabilities and market structure? The results have been staggering. I'm more patient/selective, more dynamic in my scaling, I've learnt to add into trends, my manual trading has been largely cleansed of execution errors and I have no problem taking hits. I run 3 types of manual playbook trades each producing 1/3 and they are overnight swings, a 5 min HA strategy and mean reversion (mostly short, ftse the favourite here).

Form here it becomes more about what can I take away and do less of. The systems tell me when to get long/short by watching them do their dance.

The process is this. Work out what you want/need the system to do for you and define it in 3 steps. Test an entry on multiple TF/indices by limited data (I weight recent over past which is a form of curve fitting in itself but you have to choose something), count test and 1:1 then 1:2 with spreads and brokerage.

Select the most probable results by WR to steer you in the direction of best areas to keep building. If you have found something here then you can move to testing exits in a similar manner.

If you can get a slight edge here then you can move on to using machine learning to optimise parameters for stops and targets if you have them for this step only on limited data. Then do the same for a series of filters and test combinations of these. If you can can get an equity curve of interest here then you can now open up some more data to see if it has a chance of the next stage.

Most likely by now you will have scraped 10-30 ideas. Good.

Keep doing this until something speaks to you. Now you have started your own personal database and once you have 1-300 of these maybe 5 of them you will start using across the board.

After this comes OOS and Walk Forward testing. Then finally Monte Carlo and position sizing. If you get a pass here you may have a C/B grade bot that will actually pay you something but you will also never be the same trader again.

..and finally compare your new pride to other systems for sale and see how it stacks up. Throwing a dart at NDQ long is a good place to start as everyone has one and there are plenty for sale, none of which you'd touch of course.

That's probably too much as it is now I've written a short hand book so I will take a holiday. Hopefully this is useful to more than just you. I hope I haven't gone too off track as this is a real passion for me.

GL

Expand

You are a legend.

Appreciate you taking the time for such a detailed response.

XJO Value at posting: 6,954.2
Upvote

0 Great analysis

Reply

Top Stories

ASX hits all time high out the gate on Friday trades following bullish US inflation data

MDI

Nanocap Middle Island Resources kicks off Barkly copper-gold drill run – a potential IOCG-style NT project

NIM

Nimy to raise $1.21M to explore 'newsy' WA targets

IMM

Immutep reports strong response rate from immunotherapies treating carcinomas

I88

Infini soars 50% as company reveals it could’ve just made uranium history – redefining what it means to be “high grade”

Week 27 Wrap: ASX hits all time high; CBA shares' two realities; US inflation hits 3% ahead of China plenum

BHP

BHP dips as WA nickel suspended after A$6B hit; losses eased by US CPI and forewarning

AII

Almonty locks in govt agreement for tungsten oxide plant in South Korea

Bitcoin investors need to know the risks

SKS

SKS Technologies wins US$90M award to supply power to international hyperscale data centre in Melbourne

There are more pages in this discussion • 17 more messages in this thread...

First 282 Last

Add XJO (ASX) to my watchlist