Market prediction based on macroeconomic indicators - page 8
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Replacing one straight line regression with two straight lines, one for positive input values and one for negative input values, showed no particular advantage. Attempting to predict GDP instead of S&P500 resulted in lower RMS of the predictions, but the optimal number of inputs is still 1. So adding a second and third predictor always leads to an increase in prediction RMS. I don't like this. I would like to see a model with more variables. So the search for models, data and their transformations continues.
increasing the number of variables naturally increases the overall variance
optimal number of predictors from 5 to 8 (imho)
Dow Jones and the three-month LIBOR interest rate on the dollar. Strangely, the textbooks say that when the rate goes up, the market goes down, but it's the other way round.
However, a strong correlation is not seen.
P.S. Who can tell me where to find data on interest rates before 1986 and prices (not yields) of Tregers older than 2007?
Dow Jones and the three-month LIBOR interest rate on the dollar. Strangely, the textbooks say that when the rate goes up, the market goes down, but it's the other way round.
However, a strong correlation is not seen.
P.S. Who can tell me where to find the data on interest rates before 1986 and prices (not yields) of T-bills older than 2007?
http://www.treasury.gov/resource-center/data-chart-center/interest-rates/Pages/Historic-LongTerm-Rate-Data-Visualization.aspx
but really very inconvenient (((.
http://www.treasury.gov/resource-center/data-chart-center/interest-rates/Pages/Historic-LongTerm-Rate-Data-Visualization.aspx
but it's really awkward (((
Sorry, it only has rates.
the prices are here but only by the month
http://www.investing.com/rates-bonds/us-10-yr-t-note-historical-data
Replacing a single line regression with two line regressions, one for positive input values and one for negative input values, did not show much benefit. Attempting to predict GDP instead of S&P500 resulted in lower RMS of the predictions, but the optimal number of inputs is still 1. So adding a second and third predictor always leads to an increase in the RMS of the pre-prediction. I don't like this. I would like to see a model with more variables. So the search for models, data and their transformations continues. I don't want to deceive myself by sampling predictors based on all history and predicting from the same history by selected predictors. The real challenge now is how to select predictors with only history up to the predicted date. Maybe reality really limits the choice to only one predictor.
I don't understand why you don't RAttle - you use some cheesy and, more importantly, limited models.
I have a suggestion.
Send me the tsv.file with the names of the columns. Specify which (which) columns should be used as target variables. Naturally, the table row should refer to one point in time.
I'll run it in Rattle and with your permission I'll post the result here for 6 very decent models.
PS.
Linear regressions on financial series..... more than questionable.
Sorry, it's just the rates.
The prices are here but only by the month.
http://www.investing.com/rates-bonds/us-10-yr-t-note-historical-data
Thank you! Daily data judging by the chart on the website from 2013.
Linear regressions on financial series..... more than questionable.
why, as simple models it's quite suitable
of course not as versatile as neural networks but still
I often observe that if a regression does not provide good quality models then other optimizers do not provide
Why not, as simple models are quite suitable
of course not as versatile as neural networks but still
often observed that if regression does not give a good quality model then other optimizers do not give
Yeah, well...
Regression models are almost impossible to apply to financial data. You get the numbers, you believe them, and the fact that the numbers you see don't exist at all, you lack the knowledge to understand it - a numbers game.
But my business is to suggest... And there, it's up to you...
PS.
Neural networks do not give the best results. There are other models, and the result is better and the internal structure is interpretable.
Yeah, okay...
Regression models are practically not applicable on financial data. You get the numbers, you believe them, and the fact that the numbers you see don't exist at all, you lack the knowledge to understand it - a numbers game.
But my business is to suggest... And there, it's up to you...
PS.
Neural networks do not give the best results. There are other models, and the result is better and the internal structure is interpretable.
I would still disagree - regression works fine with any data, not necessarily better than other methods, but still good enough especially when you consider the extreme undemanding nature of computational resources
It is usually recommended to logarithm or take deltas before regression - but that kills the trend information! - maybe that's why you're sceptical about regression?
pre-normalisation can spoil the data, it needs to be done very carefully
I agree that a model has to make "physical" sense... and the more complex a model, the more it is detached from "physical" interpretation, in any complex model the coefficients are abstract (unless it corresponds to lots or observation numbers/balls, multipliers to calculate volatility or something like that), neural networks are abstract, random forests are also abstract, what else? genetics? also an abstract model
In regression, coefficients are usually interpreted as the strength of correlation/influence of a factor and it is logical enough to calculate the sum of moduli of coefficients and to calculate the share of a coefficient in the sum - this will be the level of significance of influence of the variable
Of course, this cannot always be expressed in economic terms (in this case you need to build a solid model and you can trust these figures, but this is a different level), for example if you analyze relationships between the stock index and macroeconomic statistics, you will see something like "% growth of index to % growth of index" or for example there is no direct relationship between balances in accounts of the Central Bank and an exchange rate, but the model may show that there is a relationship, not necessarily an objective one (that balances somehow influence the exchange rate or vice versa) but the model shows synchronous changes, so
if the model contains only traded instruments, we can recalculate coefficients in lots - more than a physical interpretation