Non-Farm Payroll (NFP) Model Forecast – February 2015

This article presents the Trader Edge aggregate neural network model forecast for the February 2015 non-farm payroll data, which is scheduled to be released tomorrow morning at 8:30 AM EST.

Non-Farm Payroll (NFP) Model Forecast - February 2015

The Trader Edge aggregate NFP model represents the average of three neural network forecasting models, each of which employs a different neural network architecture.  Unlike expert systems, neural networks use algorithms to identify and quantify complex relationships between variables based on historical data.  All three models derive their forecasts from seven explanatory variables and the changes in those variables over time.

The table in Figure 1 below includes the monthly non-farm payroll data for two months: January and February 2015.  The January data was released last month and the non-farm payroll data for February 2015 will be released tomorrow morning at 8:30 AM EST.

The model forecasts are in the third data row of the table (in blue).  Note that past and current forecasts reflect the latest values of the independent variables, which means that forecasts will change when revisions are made to the historical economic data.

The monthly standard error of the model is approximately 76,800 jobs.  The first and last data rows of the table report the forecast plus 0.5 standard errors (in green) and the forecast minus 0.5 standard errors (in red), respectively.  All values are rounded to the nearest thousand.  If the model errors were normally distributed, roughly 31% of the observations would fall below -0.5 standard errors and another 31% of the observations would exceed +0.5 standard errors.

The actual non-farm payroll release for January 2015 is in the second data row of the table (in purple).  The consensus estimate (reported by for February 2015 is also in the second data row of the table (in purple).  The reported and consensus NFP values also include the deviation from the forecast NFP (as a multiple of the standard error of the estimate).  Finally, the last column of the table includes the estimated changes from January 2015 to February 2015.

Figure 1: Non-Farm Payroll Table February 2015

Figure 1: Non-Farm Payroll Table February 2015

Model Commentary

The aggregate neural network model forecast for February is 192,000, which is down a significant 86,000 jobs from last month's revised forecast of 278,000.  The sharp drop in the forecast from January to February reflects a surprising weakening in the employment environment during the month of February. The consensus estimate for February is 240,000, which is down 17,000 jobs from the January report, indicating a minor softening in the employment environment.  The actual January data was slightly below the revised January forecast (-0.27 S.E.) and the consensus estimate for February is notably higher than the model forecast (+0.63 S.E.).

The Government revised the NFP data last month and (not surprisingly) introduced a large new outlier in the November 2014 data, suggesting an increase of 423,000 jobs (+1.64 S.E.).  The revised December 2013 data is still an outlier (-1.72), but is smaller than their previous NFP data error (-2.03 S.E). If we ignore the two large NFP outliers, there has been a gradual and sustained positive trend in the employment data from mid-2012 through January of 2015. The trend is easier to see in the forecast data due to fewer outliers. To test the strength of the trend that began in mid-2012, last month I ran a simple linear regression on the monthly NFP forecast data (dependent variable) against time (independent variable).

The R-squared of the linear regression was 56%, indicating that 56% of the variation in the forecast data was explained by the simple linear trend model. The coefficient for the monthly slope variable was 3.12, which represented an incremental increase of 3,100 jobs per month. The slope coefficient was highly significant with a t-statistic of 6.41 and a P-value of 3.31E-07. The standard error of the model was 27,850 jobs.

The simple linear regression model cannot be used long-term due to the cyclical nature of employment, but it can be useful for quantifying short-term trends in the data. The linear regression model predicted a gain of 265,000 jobs in January, which was very close to both the revised aggregate model forecast (278,000) and actual NFP data (257,000).

The linear regression model forecast for February 2015 is 268,000 which is much higher than the aggregate model forecast of only 192,000. If the aggregate model forecast of 192,000 is accurate, it would represent a negative 2.72 standard error event. The large divergence between the short-term linear regression forecast and the more sophisticated and adaptive aggregate model forecast is troubling and may indicate a weakening in the positive employment growth trend that has been in place since mid-2012.

The notable difference between the February NFP consensus and February NFP forecast suggests a material probability of an downside surprise tomorrow. The actual December 2013 NFP data and the November 2014 NFP data were clearly outliers, but the Trader Edge model has tracked the actual NFP data very closely over the past few years.

Figure 2: Non-Farm Payroll Graph February 2015

Figure 2: Non-Farm Payroll Graph February 2015


Basic forecasting tools can help you identify unusual consensus economic estimates, which often lead to substantial surprises and market movements.  Identifying such environments in advance may help you protect your portfolio from these corrections and help you determine the optimal entry and exit points for your strategies.

In the case of the NFP data, the monthly report data is highly variable and prone to substantial revisions.  As a result, having an independent and unbiased indicator of the health of the U.S. job market is especially important.

Print and Kindle Versions of Brian Johnson's Book are Both Available on Amazon (35 5-Star Reviews)

Option Strategy Risk / Return Ratios: A Revolutionary New Approach to Optimizing, Adjusting, and Trading Any Option Income Strategy

Trader Edge Strategy E-Subscription Available: 20% ROR

The Trader Edge Asset Allocation Rotational (AAR) Strategy is a conservative, long-only, asset allocation strategy that rotates monthly among five large asset classes. The AAR strategy has generated 20%+ annual returns over the 20+ year combined back and forward test period.  Please use the above link to learn more about the AAR strategy.


Your comments, feedback, and questions are always welcome and appreciated.  Please use the comment section at the bottom of this page or send me an email.


If you found the information on www.TraderEdge.Net helpful, please pass along the link to your friends and colleagues or share the link with your social or professional networks.

The "Share / Save" button below contains links to all major social and professional networks.  If you do not see your network listed, use the down-arrow to access the entire list of networking sites.

Thank you for your support.

Brian Johnson

Copyright 2015 - Trading Insights, LLC - All Rights Reserved.


About Brian Johnson

I have been an investment professional for over 30 years. I worked as a fixed income portfolio manager, personally managing over $13 billion in assets for institutional clients. I was also the President of a financial consulting and software development firm, developing artificial intelligence based forecasting and risk management systems for institutional investment managers. I am now a full-time proprietary trader in options, futures, stocks, and ETFs using both algorithmic and discretionary trading strategies. In addition to my professional investment experience, I designed and taught courses in financial derivatives for both MBA and undergraduate business programs on a part-time basis for a number of years. I have also written four books on options and derivative strategies.
This entry was posted in Economic Indicators, Market Commentary, Market Timing, NFP Forecasting Model and tagged , , , , , , . Bookmark the permalink.

Leave a Reply

Your email address will not be published. Required fields are marked *