Actuarial Outpost

Actuarial Outpost (http://www.actuarialoutpost.com/actuarial_discussion_forum/index.php)
-   Finance - Investments (http://www.actuarialoutpost.com/actuarial_discussion_forum/forumdisplay.php?f=12)
-   -   CPI index modeling (http://www.actuarialoutpost.com/actuarial_discussion_forum/showthread.php?t=200700)

tope 09-05-2010 11:37 AM

CPI index modeling
 
Does anyone know of an excel based model that can be used to project the CPI rate as well as the health care inflation rates...I have data dating back 36 months

Im trying to figure out if ARMA, Moving average or AR models will work

which is most robust?

Colymbosathon ecplecticos 09-05-2010 12:49 PM

Why would you expect any of those models to be predictive?

What (exactly) do you mean by robust in this usage?

If you are going to bet money on this, can I be your counter-party? If not, can I at least short your stock before you do it?

tope 09-05-2010 03:57 PM

Good question

I don't think any model can be fully predictive, however we do make projections and measure the error term as well ;)

By Robust, I mean one which has enough flexibility to include known variables in the current period about the immediate future

Colymbosathon ecplecticos 09-05-2010 06:13 PM

You might look at the implied forward prices you can get from the I-bond/Treasury spreads.

bdschobel 09-05-2010 07:32 PM

I have tracked and studied the CPI for various purposes for more than 30 years. I used to buy CPI projections from commercial forecasters who claimed enormous expertise. Eventually, I concluded (based on actual experience) that nobody has any expertise in this area, and even the most expensive, sophisticated forecasts aren't worth the paper they're printed on. Even short-term forecasts (like 90 days ahead) were essentially worthless.

If anyone could forecast the CPI with any credibility at all, they'd be the richest people on the planet.

Bruce

Brad Gile 09-06-2010 12:50 AM

Quote:

Originally Posted by bdschobel (Post 4672598)
I have tracked and studied the CPI for various purposes for more than 30 years. I used to buy CPI projections from commercial forecasters who claimed enormous expertise. Eventually, I concluded (based on actual experience) that nobody has any expertise in this area, and even the most expensive, sophisticated forecasts aren't worth the paper they're printed on. Even short-term forecasts (like 90 days ahead) were essentially worthless.

If anyone could forecast the CPI with any credibility at all, they'd be the richest people on the planet.

Bruce

Indeed. There are way too many unknown variables involved. I would rather try to predict stocks over time, and that is a fool's errand unless you have inside information. I always loved building mathematical models, but some things simply are not practical to model.

campbell 09-06-2010 10:09 AM

Quote:

Originally Posted by bdschobel (Post 4672598)
I have tracked and studied the CPI for various purposes for more than 30 years. I used to buy CPI projections from commercial forecasters who claimed enormous expertise. Eventually, I concluded (based on actual experience) that nobody has any expertise in this area, and even the most expensive, sophisticated forecasts aren't worth the paper they're printed on. Even short-term forecasts (like 90 days ahead) were essentially worthless.

If anyone could forecast the CPI with any credibility at all, they'd be the richest people on the planet.

Bruce

I remember making a CPI model some years back for a lifecycle mutual fund project.... the residuals were always very large compared to my model.

Interestingly, I couldn't find any significant correlations to stuff like stock indices... and even bond indices. Perhaps it was my data set, but people were surprised when I showed them that.

bdschobel 09-06-2010 10:28 AM

I went through exactly the same process probably 20 years before you did. I showed how bad all the professional forecasts were in relation to actual experience -- something that nobody had done, apparently. And I showed that relating CPI to any other time series was basically a waste of time because correlations were so poor over any reasonable time period.

Bruce

JMO 09-07-2010 08:34 AM

Maybe this problem needs some stochastic modeling, with a high variance term, rather than an attempt to get an answer that's a single number.

oedipus rex 09-07-2010 10:50 AM

just assume its a martingale and E(X) = current value


All times are GMT -4. The time now is 08:23 AM.

Powered by vBulletin®
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

Page generated in 0.21223 seconds with 9 queries