Why are you guessing? It's all measurable.
It's clear you are in way over your head, yet expected some magic returns? Did you really think it would be that easy? You are kidding, right?
Algo & Quant traders
- Thread starter wxytrader
- Start date
Why are you guessing? It's all measurable.
It's clear you are in way over your head, yet expected some magic returns? Did you really think it would be that easy? You are kidding, right?
There are many reasons why backtest can be meaningless.
Without any details, no one can tell you what went wrong.
Hello p0box4,
Yes, I thought it would be that simple.
1. Think of trading idea.
2. Program it
3. Optimize 50% in sample 50% out of sample
4. Does it equity curve look good. Does Drawdown looks good. Net profit looks good.
5. Yes it look good.
6. Deposit money in broker and make money and hope for the best.
7. Loss $15,000
I mean, are you telling me its more complicated than that???
I guess so, I have no idea. That is why I returned my damn ass back to clicking everyday. much easier.
You didn't do 50% though did you. You did 13/15 years as optimisation. Then in two years out of sample you got lucky or perhaps there was implicit fitting because you junked the strategies with poor out of sample performance.
No it's not easy. If it was everyone could do it. But yes everyone can just randomly click and feel like they are a genius because they got lucky for a few weeks.
Like i said trading is weird.
GAT
Hello globalarbtrader,
For this algo. I optimized from 2006 to end of 2018. Then I tested out of sample from 2019 to end of November 2022.
I was happy.
Withdrew $15K from savings.
Deposit in broker. By march 2023, I was down about -$8K. Rest of the year went to about -$15K total.
I checked trades and everything. The algo just lost. The ES market just did not like it.
When I looked at the chart in ES, in 2023, breakouts failed a lot.
I got no answer. So I stopped running.
Hello globalarbtrader,
Now this you and I can agree with!!!!!
It is not easy.
When I first start building algos, nothing worked. Even things I thought would work, failed. It is very hard to build algos that work from my limited experience.
My two cents in between some back-checking of historical price data:
1) The problem with algorithmic or system trading in general is that market conditions are constantly in flux between various states or modes and degrees of volatility. This is why one algorithm can thrive periodically, but fail massively next. Simple, really.
So, you need a super-duper adaptive algorithm/system to solve this or you need multiple systems and rules for when to stop trading one and when to start trading another.
For the last few months, the "Buy the dip" system have performed very well, but it would have been crushed over the last few weeks where the "Sell the rips" system have excelled. To be fair, buying the dips have still worked, but boy, you'd need deep pockets on some of them. Just a simple example.
2) Back-testing / back-checking
Extremely important. If a system doesn't perform well historically, regardless of if that system is algorithmic or discretionary - how can you expect it to perform well in the future?
Now, fooling around in some out of the box strategy builder tweaking and optimizing indicators to fit the past doesn't sound very promising to me. I would not be surprised such an approach would fail.
I would guess a big reason why back-testing doesn't always yield satisfactory results is because the input data is usually limited to OHLC data be that minute data or tick data. And if that's so, it means a very superficial view of the market only.
The holy grail would be that adaptive algorithmic system that works well across all market conditions, but it's probably out of reach for most. Personally, I think there are things that are very hard to code and I honestly believe there's an X factor which can't be programmed and where an experienced human trader can excel.
Just common sense really. Then again, @SimpleMeLike said he doesn't know what common sense is. LOL.
1) The problem with algorithmic or system trading in general is that market conditions are constantly in flux between various states or modes and degrees of volatility. This is why one algorithm can thrive periodically, but fail massively next. Simple, really.
So, you need a super-duper adaptive algorithm/system to solve this or you need multiple systems and rules for when to stop trading one and when to start trading another.
For the last few months, the "Buy the dip" system have performed very well, but it would have been crushed over the last few weeks where the "Sell the rips" system have excelled. To be fair, buying the dips have still worked, but boy, you'd need deep pockets on some of them. Just a simple example.
2) Back-testing / back-checking
Extremely important. If a system doesn't perform well historically, regardless of if that system is algorithmic or discretionary - how can you expect it to perform well in the future?
Now, fooling around in some out of the box strategy builder tweaking and optimizing indicators to fit the past doesn't sound very promising to me. I would not be surprised such an approach would fail.
I would guess a big reason why back-testing doesn't always yield satisfactory results is because the input data is usually limited to OHLC data be that minute data or tick data. And if that's so, it means a very superficial view of the market only.
The holy grail would be that adaptive algorithmic system that works well across all market conditions, but it's probably out of reach for most. Personally, I think there are things that are very hard to code and I honestly believe there's an X factor which can't be programmed and where an experienced human trader can excel.
Just common sense really. Then again, @SimpleMeLike said he doesn't know what common sense is. LOL.
My experience, you need experience.
I build prototype fishing equipment at home and it has taken me years and 20 (could be more) odd attempts so far to build something which I still haven't perfected. Each design is often radically different from the previous.
It's a type yacht which takes a fishing line offshore.
With trading I never back test, only forward test with real money.
You've just got to experiment with your own brain in action imo.
I build prototype fishing equipment at home and it has taken me years and 20 (could be more) odd attempts so far to build something which I still haven't perfected. Each design is often radically different from the previous.
It's a type yacht which takes a fishing line offshore.
With trading I never back test, only forward test with real money.
You've just got to experiment with your own brain in action imo.
You mean coding is easy. Having a profitable strategy before you even think about coding is the hard part. Sadly, many traders (especially those that come from the programming world) thinks simply coding and backtesting will give them a profitable strategy. LOL Yeah, whatever.
Hello schizo,
My goodness you are correct. You are soo right. I thought the same thing.
Coding is easy.
Coding without an ALREADY profitable strategy is very hard.