[sgbuffett]

Last few days I have been trying replicate the results of this paper. I would say the approach is promising and has potential

https://www.sciencedirect.com/science/ar...0919302789

[sgbuffett]

I intend to expand the data set used by feed not just index stocks movement but non index stocks and open data window much bigger. Then study the predict accuracy of the forward prediction by days weeks and months.

[WhatDoYouThink?]

aiya another academic study with no practicality.

as long as you use time series which are non-stationary, no matter what models you feed them to, there's nothing to learn let alone to predict.

[Niubee]

Waste time. USA market is in correction. Most growth stocks has corrected. Only top 5 megacap stocks is holding indexes up. Apple strong... The rest is slowly down..

[sgbuffett]

aiya another academic study with no practicality.

as long as you use time series which are non-stationary, no matter what models you feed them to, there's nothing to learn let alone to predict.

That is not my understanding.

I validated it by splittingvthe data to 6 periods some for training and others for testing.

The approach is useful as alternative to other models used...
There is certainly more work to be done...i did not say it will work well enough with time series alone.

J intend to just expand the data to a crazy large set including day of the week, month, economic data ....etc so that any market inefficiency will be discovered.

[Sentinel]

That is not my understanding.

I validated it by splittingvthe data to 6 periods some for training and others for testing.

The approach is useful as alternative to other models used...
There is certainly more work to be done...i did not say it will work well enough with time series alone.

J intend to just expand the data to a crazy large set including day of the week, month, economic data ....etc so that any market inefficiency will be discovered.

So that's how you spend your cocooned time instead of making money

No wonder you're poor

[WhatDoYouThink?]

That is not my understanding.

I validated it by splittingvthe data to 6 periods some for training and others for testing.

The approach is useful as alternative to other models used...
There is certainly more work to be done...i did not say it will work well enough with time series alone.

J intend to just expand the data to a crazy large set including day of the week, month, economic data ....etc so that any market inefficiency will be discovered.

aiya so simple still cannot understand. how about these:

1. market time series are non-stationary, no matter how big are yr data sets, and how many subsets

2. models cannot be built on non-stationary data, no hidden rule, nothing to extract from, unpredictable 

nvml, you pls try and let us know your findings

[sgbuffett]

aiya so simple still cannot understand. how about these:

1. market time series are non-stationary, no matter how big are yr data sets, and how many subsets

2. models cannot be built on non-stationary data, no hidden rule, nothing to extract from, unpredictable 

nvml, you pls try and let us know your findings

If you assumption is it's a dice roll..then there is nothing to discover. However that is the question we are trying to answer...rather than the starting point.

You cannot prove non-stationarity using traditional techniques that cannot cannot uncover and discover anything.

Anyway Renaissance Tech returns tells us there are patterns to uncover.

[debster]

I intend to expand the data set used by feed not just index stocks movement but non index stocks and open data window much bigger. Then study the predict accuracy of the forward prediction by days weeks and months.

What data to add besides daily closing price and volume?

[Sharexchange]

I use Moon Phase.

[WhatDoYouThink?]

If you assumption is it's a dice roll..then there is nothing to discover. However that is the question we are trying to answer...rather than the starting point.

You cannot prove non-stationarity using traditional techniques that cannot cannot uncover and discover anything.

Anyway Renaissance Tech returns tells us there are patterns to uncover.

aiya I din say it's dice roll aka random walk. I only said the data are non-stationary. 

I aso never say there's no pattern. in fact market fractals, which are recurring geometrical patterns for determining market turning pts and directions are quite promising