General Trading Chat
- Thread starter timepass
- Start date
- Tags idle thoughts trading talk only
I understand how NIFTY Bank indexes builds up - but indices are not predicted as sum of the predictions of underlying stocks because people like you are also trading on indices itself - so it uses a different algorithm that I use. I think I could do something for derivatives....
Last edited by a moderator:
Let me explain what @Arindam Bhattacharjee is doing in layman terms so it is easy for everyone to understand. When we say machine learning, it is just that we are using computer to process some historical data and define a prediction formula. For example lets say we are trying to solve a linear regression problem, then we can have a formula to define it as follows:
Y = w0 + w1X1 + w2X2 + w3X3
Where Y is the predicted value, and X1,X2,X3 are the parameters we use to predict the value for Y, for example X1 = Average rainfall in x days, X2 = Crude oil price, X3 = S&P value etc
w1,w2,w3 are the weightages to be given to each of the parameters so that the predicted values Y closely matches to actual result.
Now what we do is give historical data set for Y,X1,X2,X3 to computer and regress it to find the best fit values for w0,w1,w2,w3 and the computer can output a best fit formula for the given historical data set something like:
Y = 0.5 + 0.4X1 + 1.5X2 + 0.7X3
Now we can use this formula to predict future values, i.e if we have a new value for X1,X2,X3 we just substitute it in the formula to get the prediction value for Y.
Thats it...not a rocket science if we understand it well...and we have a new ML model..
@Arindam Bhattacharjee , I feel it would be useful to others if you give actionable insights something like this:
NIFTY Trend for Tomorrow = UP with probability of 80%
NIFTY HIGH value for Tomorrow = 11700-11725 with a probability of 70%
Also for others to have confidence on your model you should mention how you have built your model. Dont have to give the actual trained model values as it would be your proprietary info, just mention your hypothesis and what parameters you feel are important and why.
What is the accuracy of your KNN trend model? You can share the confusion matrix.
What is the Average Error for the range prediction model? You can share the RMSE value (Root Mean Square Error)
I have build stable models which have accuracy between 55-60% but not suitable for retail trading due to draw-downs and cost overhead. The effort involved should be compensated with higher returns else no point putting money in it...so far I have found simple rule based mechanical systems are giving far better returns.
Y = w0 + w1X1 + w2X2 + w3X3
Where Y is the predicted value, and X1,X2,X3 are the parameters we use to predict the value for Y, for example X1 = Average rainfall in x days, X2 = Crude oil price, X3 = S&P value etc
w1,w2,w3 are the weightages to be given to each of the parameters so that the predicted values Y closely matches to actual result.
Now what we do is give historical data set for Y,X1,X2,X3 to computer and regress it to find the best fit values for w0,w1,w2,w3 and the computer can output a best fit formula for the given historical data set something like:
Y = 0.5 + 0.4X1 + 1.5X2 + 0.7X3
Now we can use this formula to predict future values, i.e if we have a new value for X1,X2,X3 we just substitute it in the formula to get the prediction value for Y.
Thats it...not a rocket science if we understand it well...and we have a new ML model..
@Arindam Bhattacharjee , I feel it would be useful to others if you give actionable insights something like this:
NIFTY Trend for Tomorrow = UP with probability of 80%
NIFTY HIGH value for Tomorrow = 11700-11725 with a probability of 70%
Also for others to have confidence on your model you should mention how you have built your model. Dont have to give the actual trained model values as it would be your proprietary info, just mention your hypothesis and what parameters you feel are important and why.
What is the accuracy of your KNN trend model? You can share the confusion matrix.
What is the Average Error for the range prediction model? You can share the RMSE value (Root Mean Square Error)
I have build stable models which have accuracy between 55-60% but not suitable for retail trading due to draw-downs and cost overhead. The effort involved should be compensated with higher returns else no point putting money in it...so far I have found simple rule based mechanical systems are giving far better returns.
Last edited:
Let me explain what @Arindam Bhattacharjee is doing in layman terms so it is easy for everyone to understand. When we say machine learning, it is just that we are using computer to process some historical data and define a prediction formula. For example lets say we are trying to solve a linear regression problem, then we can have a formula to define it as follows:
Y = w0 + w1X1 + w2X2 + w3X3
Where Y is the predicted value, and X1,X2,X3 are the parameters we use to predict the value for Y, for example X1 = Average rainfall in x days, X2 = Crude oil price, X3 = S&P value etc
w1,w2,w3 are the weightages to be given to each of the parameters so that the predicted values Y closely matches to actual result.
Now what we do is give historical data set for Y,X1,X2,X3 to computer and regress it to find the best fit values for w0,w1,w2,w3 and the computer can output a best fit formula for the give historical data set something like:
Y = 0.5 + 0.4X1 + 1.5X2 + 0.7X3
Now we can use this formula to predict future values, i.e if we have a new value for X1,X2,X3 we just substitute it in the formula to get the prediction value for Y.
Thats it...not a rocket science if we understand it well...and we have a new ML model..
@Arindam Bhattacharjee , I feel it would be useful to others if you give actionable insights something like this:
NIFTY Trend for Tomorrow = UP with probability of 80%
NIFTY HIGH value for Tomorrow = 11700-11725 with a probability of 70%
Also for others to have confidence on your model you should mention how you have built your model. Dont have to give the actual trained model values as it would be your proprietary info, just mention your hypothesis and what parameters you feel are important and why.
What is the accuracy of your KNN trend model? You can share the confusion matrix.
What is the Average Error for the range prediction model? You can share the RMSE value (Root Mean Square Error)
I have build stable models which have accuracy between 55-60% but not suitable for retail trading due to draw-downs and cost overhead. The effort involved should be compensated with higher returns else no point putting money in it...so far I have found simple rule based mechanical systems are giving far better returns.
Y = w0 + w1X1 + w2X2 + w3X3
Where Y is the predicted value, and X1,X2,X3 are the parameters we use to predict the value for Y, for example X1 = Average rainfall in x days, X2 = Crude oil price, X3 = S&P value etc
w1,w2,w3 are the weightages to be given to each of the parameters so that the predicted values Y closely matches to actual result.
Now what we do is give historical data set for Y,X1,X2,X3 to computer and regress it to find the best fit values for w0,w1,w2,w3 and the computer can output a best fit formula for the give historical data set something like:
Y = 0.5 + 0.4X1 + 1.5X2 + 0.7X3
Now we can use this formula to predict future values, i.e if we have a new value for X1,X2,X3 we just substitute it in the formula to get the prediction value for Y.
Thats it...not a rocket science if we understand it well...and we have a new ML model..
@Arindam Bhattacharjee , I feel it would be useful to others if you give actionable insights something like this:
NIFTY Trend for Tomorrow = UP with probability of 80%
NIFTY HIGH value for Tomorrow = 11700-11725 with a probability of 70%
Also for others to have confidence on your model you should mention how you have built your model. Dont have to give the actual trained model values as it would be your proprietary info, just mention your hypothesis and what parameters you feel are important and why.
What is the accuracy of your KNN trend model? You can share the confusion matrix.
What is the Average Error for the range prediction model? You can share the RMSE value (Root Mean Square Error)
I have build stable models which have accuracy between 55-60% but not suitable for retail trading due to draw-downs and cost overhead. The effort involved should be compensated with higher returns else no point putting money in it...so far I have found simple rule based mechanical systems are giving far better returns.
Last edited:
Reserve Bank has a reserve of $10bilion (approx). I don't know exactly whose money it is, but apparently now the government wants to get its hands on that money.
Bimal Jalan panel favours transfer of excess RBI reserves to government
1 min read . Updated: 17 Jul 2019, 01:56 PM ISTAsit Ranjan Mishra
https://www.livemint.com/industry/b...rbi-reserves-to-government-1563351402470.html
Somehow I am not enamored of this idea. I think the reserves with the RBI are the citizen's property.
Bimal Jalan panel favours transfer of excess RBI reserves to government
1 min read . Updated: 17 Jul 2019, 01:56 PM ISTAsit Ranjan Mishra
- The Bimal Jalan panel held its last meeting on Wednesday
- The RBI had constituted the panel in December last year
https://www.livemint.com/industry/b...rbi-reserves-to-government-1563351402470.html
Somehow I am not enamored of this idea. I think the reserves with the RBI are the citizen's property.
Yes, correct it is just the simple old maths...statistics has been around for a long time now and many of the formulas we use these days are based on equations defined long back. Only recently this field is becoming popular with fancy terms like Data science, Data Scientists, Machine Learning, Artificial Intelligence etc...
The reason for popularity now is due to the huge amount of data we generate these days and the computer processing power that is becoming cheaper day by day. As more and more people are working on it we see new approaches and technologies being developed based on it such as deep learning, AI applications and many research happening in different domains like medical, finance, hospitality etc
These days you may be hearing data is money and AI is the future, that is because with so much big data we can derive data driven prediction formulas. i.e we have the advantages of make the law of large numbers/data samples to build our models with higher accuracy which was not possible earlier.
The reason for popularity now is due to the huge amount of data we generate these days and the computer processing power that is becoming cheaper day by day. As more and more people are working on it we see new approaches and technologies being developed based on it such as deep learning, AI applications and many research happening in different domains like medical, finance, hospitality etc
These days you may be hearing data is money and AI is the future, that is because with so much big data we can derive data driven prediction formulas. i.e we have the advantages of make the law of large numbers/data samples to build our models with higher accuracy which was not possible earlier.
Yep and I already mentioned it here: https://www.traderji.com/community/...rivatives-indices.108053/page-21#post-1369406
Similar threads
-
My Views on the General Elections 2024
- Started by Verde
- Replies: 11
-
-
What are some general and basic price-volume action pattern one should know?
- Started by shyttrader
- Replies: 1
-
-