Econometric analysis for trading?

[mattonline]

I am fairly new to trading but I am studying a master in economics and heavily rely on econometrical analysis for my research, so I know that well.

I wanted to know if there were any models out there in which they use econometrical analysis and macroeconomic variables to help ‘predict’ market trends, i.e. within some confidence levels.
I have recently looked into Bollinger bands and at first look they seems to work so I thought about how I could capture this data (along with other data) into a regression model.

What I am thinking of is constructing a model which takes many macroeconomic variables along with intra-day variables into account, as they say markets repeats itself so it´s possible that this may work.

I was thinking of programming a system that would download this data in intervals of 5 minutes, 1 hour or even each day into an excel spreadsheet which then will be imported into an econometric package, i.e. Stata and it will update me with ´predictions´ as such, or at least I can make inference on the market given the economic data collected.

So I am essentially just brainstorming ideas for variables I could model, i.e. highs and lows of the day, 20 day moving average, Bollinger bands as these are just numbers which can be regressed.

Let me know what you think,

Thanks

 

[malaguti]

I am fairly new to trading but I am studying a master in economics and heavily rely on econometrical analysis for my research, so I know that well.

I wanted to know if there were any models out there in which they use econometrical analysis and macroeconomic variables to help ‘predict’ market trends, i.e. within some confidence levels.
I have recently looked into Bollinger bands and at first look they seems to work so I thought about how I could capture this data (along with other data) into a regression model.

What I am thinking of is constructing a model which takes many macroeconomic variables along with intra-day variables into account, as they say markets repeats itself so it´s possible that this may work.

I was thinking of programming a system that would download this data in intervals of 5 minutes, 1 hour or even each day into an excel spreadsheet which then will be imported into an econometric package, i.e. Stata and it will update me with ´predictions´ as such, or at least I can make inference on the market given the economic data collected.

So I am essentially just brainstorming ideas for variables I could model, i.e. highs and lows of the day, 20 day moving average, Bollinger bands as these are just numbers which can be regressed.

Let me know what you think,

Thanks

technical analysis has been around for some time, so there's nothing new there. economics or not, its the same thing. There are tools out there that attempt to do the same thing, you may want to look up neural nets which can be trained on multiple inputs but at the end of the day they are all based on the same mathematical data

 

[Atilla]

I am fairly new to trading but I am studying a master in economics and heavily rely on econometrical analysis for my research, so I know that well.

I wanted to know if there were any models out there in which they use econometrical analysis and macroeconomic variables to help ‘predict’ market trends, i.e. within some confidence levels.
I have recently looked into Bollinger bands and at first look they seems to work so I thought about how I could capture this data (along with other data) into a regression model.

What I am thinking of is constructing a model which takes many macroeconomic variables along with intra-day variables into account, as they say markets repeats itself so it´s possible that this may work.

I was thinking of programming a system that would download this data in intervals of 5 minutes, 1 hour or even each day into an excel spreadsheet which then will be imported into an econometric package, i.e. Stata and it will update me with ´predictions´ as such, or at least I can make inference on the market given the economic data collected.

So I am essentially just brainstorming ideas for variables I could model, i.e. highs and lows of the day, 20 day moving average, Bollinger bands as these are just numbers which can be regressed.

Let me know what you think,

Thanks

This sounds like a great idea and I certainly would be interested with the progress and your findings.

One of my papers was Econometrics and my paper was to verify the elasticity of the Ms to changes in the interest rate back in the early 80s. Milton Friedman was man of the day - a monetarist much favoured as flavour of the day. I did find a positive correllation but was pretty much led by all the debate supporting it and unable to question and challenge data etc.

If I was to write the same paper again I'd have a different approach. What worked then does not work now for one key important factor. Expectations Theory or what may be now observed as numerous Consumer Confidence indexes. As you can see despite having real zero or negative interest rates no one really wants to borrow more money when they are not feeling optimistic or unemployement and hard times are looming ahead.

Anyhow to cut a long story short I would try and keep your model very narrow and well defined as there is much noise and statistical trash (if you don't mind me saying so) to pollute and what I can only suggest as curve fitting to fit ones arguement etc.

With this approach the most simplistic one I can come up with is forecasting exchange rate fluctuations in the FX markets.

I'm sure there must be lots of models out there already that considers BoP, real and nominal interest rates v inflation etc between two pairs.

There is then the issue of picking the currency whether it is a freely floating one, Pegged (as the Swiss Franc was until recently :cheesy or in between where some banks step in once in a while.


I'd think the recent Rouble rumble should be an interesting subject matter for study and how well and quickly Elvira Nabiullina managed to bring order by resisting capital controls and choosing to raise interest rates to 17%.


Big question to ask is in such a scenario given one has sophisticated computer based algorithmic models that can tell you how much will a 1% move effect hot capital flows, how does one determine what would be a good reasonable rate to raise interest rates to?

Rates went up from 10.5 to 17% over-night? Who decides/determines this rate and how?

Anyhow, fwiw my idea would be to stick to the FX markets as it's likely to be easier to analyse with data and readilly available stats and then work up from there.

Alternative questions to ask may be;
Effect of 1% increase in inflation on currency
Effect of 1% increase in GDP on currency
Effect of 1% increase in +ve BoP

I'm sure these have all been done and models exist but to blend this in with daily currency fluctuations is a tough call.

You can look at Average Trading Ranges, BBands, Volatility, Futures and Options and obviously commodities; gold and oil in particular.

Then you have geopolitical factors.

Start out small with a well defined and controlled model / relationships and then branch out. From an economics perspective you'd need to consider whether one is a consumption based economy or productive one. Manufacturing or agricultural. Russia with oil, gas and gold obviously a commodity based one. You could look at composition of economies and how your model plays out between continents etc etc.

You might want to appraise Greece's dilemma whether to revert back to the Drachma and ask the question what would be a reasonable approach predicted by your model on determining new currency.

Perhaps apply model to the Euro with respect to a multi-divergent types of economies between South and Northern Europe. Certainly much debate interest and material is available and it may lead to some interesting findings for study.

Sorry for the long blogg but thought I'd spew mad ideas at you. Great stuff wish I was back at uni again doing something like this.

Would be delighted to follow your progress.
Best regards,

 

[globalarbtrader]

I built a model like this for a large quant fund, so I know of what I speak.

First macro data. You're unlikely to get a decent Sharpe Ratio from this. Macro data is lagged. You can predict macro numbers from market numbers; vice versa doesn't work so well. There is an announcement effect from macro data, but its short lived and you'll be competing with people who have much simpler, but faster, ways of capturing the bounce. Also when backtesting you have a lot of issues around data lags, revisions etc.

Secondly on the econometrics. Running a big regression with a lot of predictors is a recipe for overfitting. You'll need to run out of sample fitting to get realistic answers. You'll also need a big dose of robustness in your regression, something like a penalty for large deviations from equal Betas. These are tricky to get right, since you have to calibrate the penalty, which isn't easy.

It's probably simpler to do the following:

a) create a series of simple forecasting models for each effect you're trying to capture (make sure you forecast in risk adjusted space, eg Sharpe Ratio)
b) combine your forecasts in a portfolio, eithier equally weighted, or using out of sample robust portfolio optimisation (this is the approach I use).

Econometrics, if used with care, is good for things like trading baskets of stocks on which you've picked up cointegrating relationships. It's overkill and potentially dangerous for this sort of application.

I am fairly new to trading but I am studying a master in economics and heavily rely on econometrical analysis for my research, so I know that well.

I wanted to know if there were any models out there in which they use econometrical analysis and macroeconomic variables to help ‘predict’ market trends, i.e. within some confidence levels.
I have recently looked into Bollinger bands and at first look they seems to work so I thought about how I could capture this data (along with other data) into a regression model.

What I am thinking of is constructing a model which takes many macroeconomic variables along with intra-day variables into account, as they say markets repeats itself so it´s possible that this may work.

I was thinking of programming a system that would download this data in intervals of 5 minutes, 1 hour or even each day into an excel spreadsheet which then will be imported into an econometric package, i.e. Stata and it will update me with ´predictions´ as such, or at least I can make inference on the market given the economic data collected.

So I am essentially just brainstorming ideas for variables I could model, i.e. highs and lows of the day, 20 day moving average, Bollinger bands as these are just numbers which can be regressed.

Let me know what you think,

Thanks

 

[Boilersuit]

I've been meaning to make econometric analysis a much larger part of my trading.
I did a fair bit of work in the past learning Gretl and a fair few theories/approaches.
Prediction can be very good, but usually more so over longer timeframes.

Would be happy to work in a team on it, time permitting.

You could throw in interest rate spreads, inflation differentials and GDP differentials between a pair for the wider frame. . And intermarket analysis for some pairs. Seasonality is an issue too.

The shorter frame model would be more classic tech analysis orientated, including volatility.

A key concern is overfitting the model.

It might be a big job, but its rewarding. Especially if you don't like just reading fundamental reports for your steer on wkly/monthly direction.