ISM Preview

One of the most influential reports of the month will be released tomorrow morning, the ISM Manufacturing Index. The consensus estimate is 52.0, with the range of estimates from 51.0 to 55.0. Last month the ISM fell from 60.4 to 53.5, with the consensus estimate at 57.5 -- a huge miss -- and the markets reacted very negatively. I made a lot of money that day shorting into the report based on Bill McBride from Calculate Risk's very nice monthly preview using the regional Fed manufacturing surveys. Here is his chart for this month.

So will the ISM miss again? When the NY and Philly Fed surveys came out, it sure looked like it. It even looked like we would see a sub-50 print, signaling a contraction in manufacturing.  The more recent surveys, and in particular the Chicago PMI, have not been nearly as weak.

For this month, I'm attempting to improve upon the accuracy of Bill's chart by using linear regression to weight the regional indexes. I also made a second model using the Chicago PMI number.

Here's the regression summary for the model using the regional Fed surveys. This month the model is estimating the ISM will be 50.38, in line with Bill's chart.

Model Summary
Model	R	R Square	Adjusted R Square	Std. Error of the Estimate
1	.930a	.866	.854	2.56350
a. Predictors: (Constant), KC, VA, NY, PH, TX

Coefficientsa
Model		Unstandardized Coefficients		Standardized Coefficients	t	Sig.
		B	Std. Error	Beta
1	(Constant)	51.518	.592		87.012	.000
	NY	.033	.031	.085	1.061	.293
	PH	.099	.035	.271	2.835	.006
	TX	.082	.027	.296	3.013	.004
	VA	.117	.028	.300	4.161	.000
	KC	.044	.041	.094	1.063	.292
a. Dependent Variable: ISM

And the model including the Chicago PMI, which has been considerably more accurate. The KC Fed survey was discarded because it was shown to be an insignificant variable, and the PH survey was discarded because of multicollinearity. I also dropped the number of observations to 30, because that's about the minimum number you need to use a normal distribution and in a time series the most recent data is the most important. Also, two and a half years worth of data coincides with the beginning of the recovery in manufacturing. This month the model is estimating the ISM will be 52.72.

Model Summary
Model	R	R Square	Adjusted R Square	Std. Error of the Estimate
1	.987a	.975	.971	1.55420
a. Predictors: (Constant), CH, VA, NY, TX

Coefficientsa
Model		Unstandardized Coefficients		Standardized Coefficients	t	Sig.
		B	Std. Error	Beta
1	(Constant)	35.029	3.283		10.669	.000
	NY	.090	.028	.194	3.213	.003
	TX	.085	.035	.240	2.402	.023
	VA	.067	.023	.168	2.979	.006
	CH	.322	.055	.452	5.897	.000
a. Dependent Variable: ISM

The estimations for both models versus the actual ISM number can be seen visually in this chart.

And for fun, let's test some critical values.

	Fed Surveys		Fed and Chicago PMI
ISM	z	p	t	p
< 50	-0.1493	0.4404	-1.7473	0.0460
< 51	0.2408	0.5952	-1.1038	0.1397
< 52	0.6309	0.7360	-0.4604	0.3244
> 52	0.6309	0.2640	-0.4604	0.6756
> 55	1.8012	0.0359	1.4698	0.0766

It will be interesting to see how well the model works tomorrow. The most recent data is showing that an ISM miss is not likely. Overall, it looks to be a boring report in line with expectations, and will probably not have much of an effect on the market.