what can be predicted and what cannot be in the market?

[Bad_Badness]

Other than death and taxes - what ?

Fecundity's arguments such as:

All things are possible
The above is itself possible
Therefore all things are possible.

I.e. Those things which are vacuous or have no value.

But to the OP. There are many who say react, don't predict. Predict here meaning, acting on what you think will happen. React here being, waiting until it happens and then acting.

It is a complex subject, and I am not really willing to try to explain all the nuances. Maybe someone else can.

 

[easymon1]

Actually, "probability" was someone else's choice word. I used it when I responded to their topics.

If the model is truly predictive, and there is no 'concept drift,' then there should be no degradation.

If the model is purely just (over)fitted, then there should be instant degradation.

If the model is truly predictive, but there is concept/model drift, then there will be degradation. The rate of degradation depends on whatever factors cause the drift, known and unknown; the calculus of those factors; and how well and often the system retrains/adapts the model to the drift, if at all.

Wow!
Thanks

What pedestrian methodology might exist to retard the rate of degredation?

 

[userque]

Wow!
Thanks

What pedestrian methodology might exist to retard the rate of degredation?

Automated or manual re-training/re-tuning
Sliding window algos
Increasing inputs, training examples, model complexity
Improving validation
Ensemble learning
Multiple-model machine learning

 

[easymon1]

Automated or manual re-training/re-tuning
Sliding window algos
Increasing inputs, training examples, model complexity
Improving validation
Ensemble learning
Multiple-model machine learning

Krikey
How do you know all this good stuff?

 

[userque]

Krikey
How do you know all this good stuff?

At the risk of sounding narcissistic:
Some things I discovered on my own.
Most things I learned via reading.
A few things I learned via videos.

 

[virtusa]

Fecundity's arguments such as:

All things are possible
The above is itself possible
Therefore all things are possible.

I.e. Those things which are vacuous or have no value.

But to the OP. There are many who say react, don't predict. Predict here meaning, acting on what you think will happen. React here being, waiting until it happens and then acting.

It is a complex subject, and I am not really willing to try to explain all the nuances. Maybe someone else can.

If all things are possible, impossibility is possible too.
Which is a contradiction as it would mean that possible and impossible are possible and impossible at the same time.
But something cannot be possible and impossible at the same time. But because everything is possible that would be possible too.

 

[deaddog]

If all things are possible, impossibility is possible too.
Which is a contradiction as it would mean that possible and impossible are possible and impossible at the same time.
But something cannot be possible and impossible at the same time. But because everything is possible that would be possible too.

 

[Pekelo]

What can be predicted: market sell off.
What can not be predicted: when it happens

 

[virtusa]

What can be predicted: market sell off.
What can not be predicted: when it happens

Prediction should not be taken as "sure to happen". It is rather an increased probability that something can happen.
My trading system "predicts" all the time what will happen. The result is that I have a higher win rate than without the "predictions". But predictions are sometimes wrong too.

 

[easymon1]

At the risk of sounding narcissistic:
Some things I discovered on my own.
Most things I learned via reading.
A few things I learned via videos.

Automated or manual re-training/re-tuning
Sliding window algos
Increasing inputs, training examples, model complexity
Improving validation
Ensemble learning
Multiple-model machine learning

If the model is truly predictive, and there is no 'concept drift,' then there should be no degradation. If the model is purely just (over)fitted, then there should be instant degradation. If the model is truly predictive, but there is concept/model drift, then there will be degradation. The rate of degradation depends on whatever factors cause the drift, known and unknown; the calculus of those factors; and how well and often the system retrains/adapts the model to the drift, if at all.

Who's got the chops to DO this stuff?
Where are they Found?