When a model becomes extinct

[Scataphagos]

Are you saying that PA and TA cannot be automated?

Conceptually, seems it could be. After all, as "they" can land a vehicle on an asteroid beyond Saturn 5 years, 4 months, 3 weeks, 5 days, 2 hours and 6 minutes from now (after doing a "gravity assist slingshot fly by" of JUPITER)... I have to believe it would be possible. Unlikely though. Would be VERY complex to automate. Not enough money in it or enough intellectual interest to induce someone to tackle the task.

Though automating Price TA would be very difficult, the application of its principles are simple... as in KISS, baby!

 

[SimpleMeLike]

If one could automate the correct application of Price TA, it would be the TRUE Holy Grail.

Hello Scataphagos,

I am definitely trying in automation PA concepts.

I like your confidence when it comes to Price TA. It gives me hope.

 

[Scataphagos]

Hello Scataphagos,

If there was a "correct" application of Price TA, the market would not exist. I just do not see edge in Price TA in my experience of trading yet. Too much discretionary and decision making. Only time will tell.

Disagree about market "would not exist".

LOL about not seeing an edge in Price TA. That's just because you haven't yet recognized the correlations. If I sat you down and pointed them out, YOU'D SHIT YOUR DRAWERS... about how they have been right there in front of your eyes every day (for years?)... and you never caught onto them! Classic example of not being able to see the forest for the trees.

Don't feel bad. Apparently few get it.... mostly because they don't know to look for them. One of my favorite quotes comes from Sherlock Holmes. When asked, "just what is it that you do Mr. Holmes"? He replied, "I observe and deduce".

And you don't have to get them all... or even most of them. If you could just pick off the low-hanging fruit and exercise stop discipline, you'd do great.

 

[Scataphagos]

Conceptually, seems it could be. After all, as "they" can land a vehicle on an asteroid beyond Saturn 5 years, 4 months, 3 weeks, 5 days, 2 hours and 6 minutes from now (after doing a "gravity assist slingshot fly by" of JUPITER)... I have to believe it would be possible. Unlikely though. Would be VERY complex to automate. Not enough money in it or enough intellectual interest to induce someone to tackle the task.

Though automating Price TA would be very difficult, the application of its principles are simple... as in KISS, baby!

Just for kicks, I wrote down off the top of my head the types of correlations that come to mind. I listed 30 of them. (There are probably a few more I'm aware of that didn't come immediately to mind... and surely some I haven't discovered.)

With at least "30 things" to watch for, it would be very difficult to program. Not only that, and while they have structure and form... you can't tell in advance where on the chart they will present for most of them.

Sounds awfully complex, doesn't it? To automate, yes. To learn the correlations and watch for them, no. Simple... KISS simple... but busy.

Correlation = "What usually comes after ______"? That would be, "what usually comes after is market rises or declines". The correlation is what fits into the "_________" blank.

 

[rb7]

Just for kicks, I wrote down off the top of my head the types of correlations that come to mind. I listed 30 of them. (There are probably a few more I'm aware of that didn't come immediately to mind... and surely some I haven't discovered.)

With at least "30 things" to watch for, it would be very difficult to program. Not only that, and while they have structure and form... you can't tell in advance where on the chart they will present for most of them.

Sounds awfully complex, doesn't it? To automate, yes. To learn the correlations and watch for them, no. Simple... KISS simple... but busy.

Correlation = "What usually comes after ______"? That would be, "what usually comes after is market rises or declines". The correlation is what fits into the "_________" blank.

If it there is no emotions and feelings, then it can be automated.
Programming a module to land on Mars is way more complicated than that...

But for sure, it ain't for the faint of heart!

 

[Scataphagos]

Programming a module to land on Mars is way more complicated than that...

I don't know that it is. I'm a science guy and accustomed to thinking that way.

With a Mars landing, you have known variables which can be calculated and projected. With the correlations in TA forming "anywhere on the chart", you don't have known values from which to start or where they will complete... so you can't(?) put them into an equation. Even if it is somehow possible, seems to be waaaaayyy difficult.

 

[remoteControl]

But the question is after how long of a time..one would give up on the model after it had such a long time of consistency. Outsiders say that it is normal that it is not performing because it is the PANDEMIC, but i dont really understand how and why pandemic has changed trading. MY model is purely based on price action not looking at volume or any other indicators.

when it has exceeded 1.x times its worst performance in history. the x is determined by your risk tolerance for running a failed system, dependent on many factors. in your case since it was a good system for so long, it would make sense to give it more time than a system you just started trading but has already reached it's max drawdown / longest losing streak (use your metric of choice) from your backtest.

 

[Zortauk]

Very good question. Noone makes money in price action trading.

There is absolutely zero proof any where on earth of any price action trader making money year to year or even beating the SP500 index return. Zero. None.

As you can see from my profile name..i am not active in these forums. I just wrote here..to see if i can meet or read other traders with similar experience. AS i said, i wrote a model..which only trades during earnings season..and has made money for 12-15 years( it had a huge run..it is not couple of years)...without a significant down quarter until PAndemic,, I would say..most of the model depends on price action( i just say most..because i also use daily ATR and average daily Volume). A friend of mine suggested, VIX has gone up significantly since pandemic. Earnings season has started, and the vix is really lowest since pandemic and i am trying to be cautiously optimistic, but the last year has crushed my confidence.

 

[syammondsandrew]

Have had many similar experiences in the past where a strategy that was minting greens suddenly went downhill. I eventually had to let go of it and turned out it was a good decision as a trader. I understand it’s very difficult, but the bottom line is how long can you sustain a losing streak.

That’s why a lot of emphasis is laid on emotional management. It’s anyways important you keep revamping your strategies, leave aside sticking with one for ages.

 

[Canoe007]

...types of correlations that come to mind. ...
With at least "30 things" to watch for, it would be very difficult to program. Not only that, and while they have structure and form... you can't tell in advance where on the chart they will present for most of them.
Sounds awfully complex, doesn't it? To automate, yes. To learn the correlations and watch for them, no. Simple... KISS simple... but busy.
Correlation = "What usually comes after ______"? That would be, "what usually comes after is market rises or declines". The correlation is what fits into the "_________" blank.

I see correlations on Price charts, seem so simple, then I struggle to define them or their components mathematically. That has resulted in a collection of TAs and custom indicators that I can individually click on/off, to include/exclude in the automatic trading decision, as the current flavour of price action changes (be nice to have my code detect those flavours, to automate that too). Some items I 'duplicate', so Item-A may become Item-A-Entry and Item-A-Exit, so they can be included/excluded for Entry or Exit separately.

In practical terms, it's a hybrid process. When I visually recognize a correct setup, I enable/disable the applicable items to configure for that setup, then enable Auto-Entry, and let it handle it as/if the setup continues to an entry point, from which it handles the position. I have the ability to manual increase/decrease an auto-running position, but I rarely do that. I have defaults for order types, but can also click Enable/Disable for the type of Orders to use for Entry, Exit, Stops, etc..

GUI wise, they're a list down the side of the screen, with the running strategy having its standard items enabled/disabled. If I tune things different than the strategy defaults, the item is highlighted as Non-Default, so I don't forget. A single click will reset to Strategy Default; another click will go back to the tuned selection.

YRMV - this has made me very adept at curve fitting when backtesting.

If the Price Action is doing something I don't recognize, then "none of the above" has been 'validated' against that, so I disable Auto-Entry and watch to learn the new flavour while it collects data.

Currently rebuilding my control structure, to accommodate additional order types and "perceived" (well, suspected) layers in behaviour...

Instead of running on my 'intended correlations', I've often wondered if having their signals (and underlying indicator values?) fed into an AI might be better at establishing practical working correlations between price action and the items I construct.