How many of you need to learn coding to work with data ?

[cruisecontrol]

analyzing the market doe snot require coding necessarily, I disagree on that.
Can you go into deep more on your overfitting concerns ? intrigued to hear abt it

For example, earlier you said:

if I give you bulk of data can you pull all stocks that had earnings on the day, gapped 30 %, spiked 10% in the first 30 min of mkt open had a low price in the first 30 min which is lower than the low price in first 60 mins, broke out through the 200 MA, below VWAP, etc etc, all in a few seconds ?

These kinds of queries are overfit city. You tweak enough filters and it looks like you're printing money in the backtest, but it's nonsense without any predictive power. If your customers don't understand this they are going to be pissed at you when they lose money. You are explicitly targeting non technical people who can't code, and thus less likely to be statistically literate. I think if you offer up such overfitting power you have a responsibility to inform people about the dangers of this.

 

[d08]

I would suggest not to “marry” to one broker API. Use platform like NinjaTrader that can connect to verious brokers. They already have plumbing done and you just need to concentrate effort on your strategy.

And if NinjaTrader sells itself to a broker or just changes policy for whatever reason? Best option is to not tie yourself to anything besides a language.

 

[Spikeet]

For example, earlier you said:


These kinds of queries are overfit city. You tweak enough filters and it looks like you're printing money in the backtest, but it's nonsense without any predictive power. If your customers don't understand this they are going to be pissed at you when they lose money. You are explicitly targeting non technical people who can't code, and thus less likely to be statistically literate. I think if you offer up such overfitting power you have a responsibility to inform people about the dangers of this.

you didnt provide any examples or specifics.
The example above is an example for a strategy that is based on a criteria, for exmaple trading only cmps below a $500M mkt-cap. how is that overfitting ?
It's like saying "I only trade AAPL"

 

[cruisecontrol]

you didnt provide any examples or specifics.
The example above is an example for a strategy that is based on a criteria, for exmaple trading only cmps below a $500M mkt-cap. how is that overfitting ?
It's like saying "I only trade AAPL"

There are an infinite number of combinations of moving averages, vwap, spike percentages and all the other things you mentioned. If you search through these combinations, some of them are going to have amazingly good backtests just by chance even though there is no real actual edge. That's called overfitting.

Knowing how to avoid overfitting is not easy. Many will fall into the trap, especially if they are statistically naive, and if it is encouraged by the design of the tool.

 

[d08]

Couldn't agree more. You really need a profitable strategy to begin with. No amount of coding experience will somehow magically give you a profitable strategy (unless you're overfitting).

It can be the other way, coding skills can help you discover profitable strats.

 

[schizo]

It can be the other way, coding skills can help you discover profitable strats.

That's called "optimization", but it's not a profitable strategy unto itself.

 

[d08]

That's called "optimization", but it's not a profitable strategy unto itself.

How would you know if a strategy is profitable in the first place? You can discover through data, it doesn't have to be just optimization.

 

[schizo]

How would you know if a strategy is profitable in the first place? You can discover through data, it doesn't have to be just optimization.

One that isn't curve fitted, which most optimized strats are IMO. However, I'll grant that codes can help speed up the process of finding one. But my bone of contention is that many think that coding is synonymous with trading. Running a bunch of scripts won't magically give you a money-making ATM.

 

[d08]

One that isn't curve fitted, which most optimized strats are IMO. However, I'll grant that codes can help speed up the process of finding one. But my bone of contention is that many think that coding is synonymous with trading. Running a bunch of scripts won't magically give you a money-making ATM.

From what I've understood, that's not far off what Renaissance is doing. Granted, they use a scientifically rigorous method of testing most likely.
I mostly agree that programmers that go into trading thinking they'll be millionaires quick due to their present skills are misled.

 

[virtusa]

My comment many trade without programming has nothing to with just ET traders. I have no idea how many ET traders use or don't use programming because there is no way for anyone to know.

You don't know either how many outside of ET program or not.

So the conclusion is that neither me or you know anything to make a profound analysis and a well argumented conclusion.

To me: the moment you use a PC you need programming skills. Maybe on a low level but even in Excel you have to create formulas to get results. So you need knowledge of the syntaxes you can use and the logic on how to build that formula.

The only exception is just drawing lines thru tops and bottoms. But if you use Gann angles, retracement percentages or Fibonacci you are already using code.