More LEI

Monday, April 23. 2007

Real GDP vs Index of Leading Economic Indicators (lagged one quarter), year-over-year percentage change, 1991-2007

As noted yesterday, the smart folks at Northern Trust are believers in the index of leading indicators. So, I ran a braindead linear regression on the relationship suggested, year-over-year GDP vs LEI, one-quarter-lagged. The relationship turns out, as suggested, to be pretty strong.

            Estimate Std. Error t value Pr(>|t|)
(Intercept)    1.986      0.231     8.6    4e-12 ***
gd1.LEIyoy     0.312      0.054     5.8    3e-07 ***

Multiple R-Squared: 0.36,       Adjusted R-squared: 0.34
F-statistic:  34 on 1 and 61 DF,  p-value: 2.5e-07

A quick chart of fitted values shows the strength and weaknesses of this naive curvefitting:

Real GDP vs regression fit and two-quarter prediction on leading indicators lagged one quarter, year-over-year percentage change, 1991-2007

I would hesitate to predict the actual level of GDP growth off of this model. However, I am comfortable with the rough predictions the last few months of leading indicators are making, of below-trend GDP growth for the current and next quarter.

I warn, that conforms with my current thesis. It is always easiest to believe results that are talking your book.

Hat tips: Asha Bangalore, Paul Kasriel

Posted by Wcw in Investing at 17:48 | Comments (6) | Trackbacks (2)

Trackbacks

Trackback specific URI for this entry

More NT curvefitting
Kasriel Recession Warning Index, combination of year-over-year percent change in real monetary base and four-quarter moving average percentage point spread between Treasury 10-yr. yield and fed funds rate, 1960-2007 To refresh your memory, the combinatio

Weblog: Wcw
Tracked: Apr 24, 17:12

Even more curvefitting
Real GDP vs 1,1-GARCH fit on CFNAI-M3 (1Q lag), OECD CLI (1Q lag) and KRWI (unlagged), 1967-2007 (1987-2007 shown) Given the mild interest in my previous runs, both the braindead OLS vs the Conference Board's leading indicators and the 1,1-GARCH fit on

Weblog: Wcw
Tracked: Apr 27, 18:09

Comments

Display comments as (Linear | Threaded)

except in 1967 !!! growth went down but no recession

#1 miju on 2007-04-24 07:46 (Reply)

I don't think that's the only exception. 1995 saw the leading indicators dip as well. The r-squared even for this, as-noted 'braindead' OLS regression is only 0.3. Don't expect too much from it.

#1.1 Wcw (Homepage) on 2007-04-24 16:17 (Reply)

Are you using a data series with "as published" data or one based upon the revisions to the LEI? (not just the 2005 benchmark changes, but those resulting from updates to the underlying data.)

Thanks -

Jeff

#2 Jeff (Homepage) on 2007-04-25 15:26 (Reply)

No idea; I refuse on principle to give the Conference Board my hard-earned cash for government data. I got the data from a State of Ohio piece. I assume it is the revised series.

What I care about for an analysis like this is direction and magnitude, nothing exact. It is an OLS regression, after all, with no effort whatever made to correct for autocorrelation or heteroskedasticity. Right now, the LEI is interesting because it's pointing down more than other indicators, and it's not useless.

For pure fit, I prefer the CFNAI. It is even more of a black box, and I have no idea where I'd get as-published data for it, but its methodology is objective and it takes a lot of data series as inputs, not just a few.

#2.1 Wcw (Homepage) on 2007-04-25 16:29 (Reply)

Thx wcw.

We encounter this all of the time in revised series. There is a some element of "look forward" in the revisions. We never know whether the "direction and magnitude" were there in real time.

I hit this in my employment report analysis as well. It is very hard to do analysis with the data that were available at the time. Too bad that sources do not record and publish that data as a series along with the revisions.

I think that the Conference Board and some economists have done some work on this.

thanks again,

Jeff

#2.1.1 Jeff (Homepage) on 2007-04-26 04:44 (Reply)

I had always gone on the assumption that initial estimates were wrong, but unbiased. On this braindead OLS, I was more worried about autocorrelation, not lookahead. Does the look-forward effect systematically affect estimates, or does it simply give you inappropriately small standard errors?

#2.1.1.1 Wcw (Homepage) on 2007-04-26 14:39 (Reply)

Add Comment