Withholding Taxes Chart
Current through the Daily Treasury Statement dated May 08, 2008.
Note: 2008 is a year-to-date comparision between the first 90 data points of 2008 and 2007.
Note: 2Q08 is calculated with the first 28 data points of 2Q08 and 2Q07. This
number will bounce around less as we move through the quarter.
If you want the data series to make your own charts, PayPal me $100 US, and I will send it to you in your preferred format.
If you want to follow the second quarter data day-by-day, I have a chart and spreadsheet on this page.
Why the Charts Above Differ
Each dot on the top chart is a full year-over-year comparison calculated for the year ending on that day. The 2008 bar on the second chart is a year-to-date comparison, which covers a shorter period than any of the dots on the top chart. While the economy is decelerating, the 2008 bar will always look much worse than the recent dots on the top chart because each one of those dots contains data from a period prior to where the deceleration began. The same is true for the 2Q08 bar on the third chart - the quarterly growth rate will be worse than the yearly growth rate because it spans a much shorter time period.
Current Situation
Lots of people are being thrown out of work and having their hours cut. The recession is deepening. While unemployment claims eased-up in mid April, it is likely to turn out to be only a blip. We may have had fewer layoffs, but lots more hour-cuts. That would explain why the withholding taxes continue to move down. And of course, the withholding numbers are not “messaged” at all like the unemployment numbers, so these are the numbers that I trust the most. (A few days after writing this paragraph, the productivity report was released on May 7, 2008 confirming what I had suspected: lots of workers had their hours cut in the first quarter.)
What is This?
Each business day, the United States Treasury Department publishes a report on the workings of the government’s income and spending. One of the items on this report is the “Withheld Income and Employment Taxes.” (Here in the USA, workers have taxes automatically deducted from each paycheck.) This project charts this data. If you are not clear on how these charts work, take a look at this page.
Why is this Data Important?
This data gives a very good look at how the US economy is doing. The more people working, and the more they are being paid, the more money the Treasury rakes in. If workers get their hours cut, get their pay cut, or get laid off, this is the fastest way to find out about it. This is “real time” data, not data that is delayed for weeks as in other reports.
As you can see from the chart below, this data accurately pin-pointed the bottom of the recession in 2002, and the turn in the stock market.
The S&P 500 set its low in October 2002, which was only after the speed of the decline plateaued. Withholding taxes were still falling at about a 3% rate, which is pretty bad. However, the market saw that the plunge had stabilized, and that was the end of the bear market. We also got the last weak GDP number in the fourth quarter of 2002. So, I think this is a very good indicator to watch as we try to spot the end to the current slowdown.
You can check my data by entering a year or month into the form above, and clicking the Submit button.
The total for that period will be calculated from the daily totals and shown above when the page reloads. Then you can pull up the appropriate report and see if it matches.
Examples:
2008 for the entire year.
2008-03 for March of 2008.
Remember that the federal fiscal year begins in October. So, to find the annual total for 2007 for example, you need to pull up the report for the last day of September 2007. Also, keep in mind that the Treasury’s numbers don’t ’round to totals’, so totals are often off by a little bit.
About the Data
Note: All of my charts use nominal dollars so far.
The government publishes this data promptly at 4pm Eastern time each business day - the same time that the stock market closes in New York City. I am usually at my computer then, and plug-in the new number.
Charting the data using the daily totals, or a moving average, doesn’t really bring out the value. We are trying to pinpoint turns in the economy, so we need to look at the growth rate. Each dot on my daily charts represent two years of digested data. The number is calculated like this: All of the withholding-tax revenue over the past year is totaled. Then, another total is calculated for the prior year, and the two numbers are compared to see what the percent change was.
For example, the number for December 31, 2007 is 5.43%. That means that the total taxes collected in 2007 were that much higher than 2006. The next day’s number would be calculated with both sets of totals shifted forward by one day. In order to create this kind of chart, you need to have daily totals, rather than just monthly totals, etc.
While I’m sure that this data-series is available on professional investing systems like the Bloomberg, I don’t have one of those and could not find it anywhere on the web. So, I spent a few hours harvesting it myself.
First, I wrote a customized web-crawler in REALbasic to download all the reports, and pull out the numbers. The data was stored in a SQLite database on my Windows Vista Ultimate system.
Then, I checked the monthly and annual totals by hand. Totals were calculated with SQL queries, and then I manually pulled up the reports for the last day of the month, and last day of the year, and inspected the totals. Where the crawler failed, I acquired the number by hand.
Since the crawler would never be used again, I did not attempt to perfect it. Parsing PDF files is not easy - especially when the format is not consistent. There are text files available now, but the old data is almost all in PDF format.
Once the data was collected, I used SQLiteManager to export each year to a CSV file. Then I used phpMyAdmin’s import feature to upload the CSV files into the MySQL database here on trivisonno.com. I then did SQL queries on the MySQL data to check the annual totals to make sure that everything uploaded correctly.
Next, I wrote a PHP program to calculate the yearly total, and yearly growth percentage for each row in the table. Of course, not every day is a business day. So, if one-year-ago from a day was a weekend or holiday, the code moves forward to the next business day to calculate the year-over-year percentage. So, the data bounces around a bit more than you would expect. There are about 250 business days in a calendar year, but each data point may have a day or two more or less than that depending on how the weekends and holidays fall. I don’t think this makes a material difference.
The code to generate the data for the first four charts is written in PHP, and the charts are drawn by the Google Chart API. The data is downloaded to your browser, then your browser sends it over to Google, and Google sends back the chart. The daily chart above the comments section is just a JPEG that I made with Excel - it does not update automatically.
See my Site Technology page for a discussion of the WordPress plugin that I wrote to handle various parts of this project.
The Financial Management Service website has this data back to 1998, though there are some problems. Here is a list by fiscal year:
1998 - Not used. There are huge gaps in the data, so this series begins effectively with fiscal year 1999 in October 1998.
1999 - Clean.
2000 - Reports are missing for 1999-12-07, 1999-12-29, and 1999-12-30. The monthly total was short by 14,684 so I divided by 3 and entered the three missing dates with a value of 4,895 each. (The report on the last day of each month shows the total for that month.)
2001 - Clean.
2002 - Clean.
2003 - Clean.
2004 - The 2004-03-26 report is missing. I used the difference between the running-totals of the prior report and the next report to calculate the missing value.
2005 - Clean.
2006 - Clean.
2007 - The 2006-10-23 report is missing. I used the difference between the running-totals of the prior report and the next report to calculate the missing value.
2008 - So far, so good.
The FMS also issues revised reports to correct errors. I used the numbers from those reports wherever they were issued.
In February 2008, the Bureau of Economic Analysis reported:
I have not adjusted the January data to account for this. First, it is a small amount. Second, if inflation were pushing some workers into higher tax brackets at the end of 2007, the Treasury would have raked in at least an equal amount of extra cash. Third, the year-over-year comparison to January 2007 would still be negative - use the “Spot-Check My Data” form above to pull up both of the January numbers and see for yourself. And lastly, I have not yet located references to support this, but I’ll bet this sort of indexation is routine considering that we live with perpetual inflation.
You will notice that my charts start in the fall of 2000, rather than 1998. That’s because you need two full years of data before you can begin calculating year-over-year percentages.
Daily Chart
This chart shows each business-day’s revenue. Our government has a very steady gig. The monster days where the Treasury can take in as much as $20-$30 billion come on every other Friday since most people are paid every two weeks:
March 24th, 2008 at 6:40 am
This is great information and I thank you for all the work you did to produce the long term chart. With respect to the early years of the decade, however, wouldn’t the significant tax cuts that were enacted impact the data more than underlying economic factors for a period of time? Similarly, as an employer, I have to adjust withholding downward each January when tax brackets are adjusted for inflation. Looking at the long-term chart, however, I don’t see downward blips each January. Is there some behind the curtain adjustment built in for this (by you or Treasury)?
March 24th, 2008 at 8:58 am
Jack,
I haven’t made any adjustments to the Treasury’s data. I don’t think they make any adjustments either because they are giving an account of their financial position rather than publishing an economic analysis.
One could make an argument for adjusting the data upward to account for tax cuts, and downward to account for inflation. I prefer to simply see what the Treasury is raking in.
I did research the tax cuts made during the last recession, but I couldn’t find any historical data accurate enough to use. The tax system is a big ball of confusion even for the experts. Trying to un-ravel the ball of confusion from several years ago would be a big job.
One import fact I found was that those tax cuts were phased-in gradually, with a “speed up” in 2003.
Matt
March 24th, 2008 at 9:13 am
Great data, very impressive. I’m sure I’ll be linking to your site in a future story or two. While the data certainly shows we are in a slowdown, we are a long ways from a recession, which really surprises me. I’ve been saying we’re in a recession since about October of 2007. I’m betting we’d need to be at about the rate of inflation or lower and of course hit negative territory eventually to say we are 100% in recession.
March 24th, 2008 at 12:28 pm
Michael,
Well, an “official” recession is based on GDP data, which is a much different animal. I like to think of this data as answering the question: “Is America hiring or firing?” The USA seems to be firing, and at an increasing pace.
And yes, it is surprising that the economy is holding up as well as it is. I just wrote a post about why I think that is:
http://www.trivisonno.com/a-different-kind-of-recession
Thanks for the kind words,
Matt
March 24th, 2008 at 10:36 pm
Your data could certainly be consistent with recession as the recession in 2001 started in April of that year and based on your chart the tax withholding was still positive. So we could be in recession and that is consistent with Warren Buffet’s analysis and ECRI etc
Excellent work on the chart.
March 25th, 2008 at 7:08 am
Thanks for the reply. Now that I think about it, the fact that employers should be adjusting withholding the first payday in January and the inflation adjustment to tax brackets has been fairly small the past few years, year over year comparisons pretty much self-compensate for this factor.
IMHO, the data in the early years of this decade seem to reflect a combination of tax cuts and the macroeconomic picture.
Jack
March 25th, 2008 at 10:27 am
Really very well done. Like you, I had gone back and manually entered all of the data to generate the daily data series going all the way to the inception of the Dept. Treasury’s data. Also like you, I had to make some modest assumptions (generally trend) to fill in the missing gaps. However, my data pattern looks a lot different than yours in the last 6 months. I’d be keen to talk with you about it if you have the interest. My email address is listed with this message. Cheers.
March 25th, 2008 at 12:05 pm
Samuel,
Thank you for the kind words, and for reminding me how I saved Warren Buffet’s life:
http://www.trivisonno.com/how-i-saved-warren-buffets-life
Matt
March 25th, 2008 at 12:33 pm
Jack,
It looks like the majority of the plunge in 2000 and 2001 happened before the tax cuts went into affect:
Economic Growth and Tax Relief Reconciliation Act of 2001.
Jobs and Growth Tax Relief Reconciliation Act of 2003
If we get more tax cuts this time around, it will probably be the same story since it takes the government so long to respond. But we will have to watch such an event very closely since it is likely to make the decline look worse than it really is.
Matt
March 25th, 2008 at 12:40 pm
Edward,
Thank you for the kind words.
What kind of chart are you calculating?
There are a lot of people interested in this data, so why don’t we talk about it here?
Matt
March 28th, 2008 at 8:21 am
Very interesting use of the Treasury Statement data!
I was going to do the following, but have not had the time — You might want to consider it:
enter the daily data into Xcel and do a least square fit using a polynomial equation (under the options tab). Pick a power of 2 for the first try. Get the equation and do a projection for the next 30 days of so. You also could take the derivative and look for a minimum. I’ve often done this with the S&P and it does have some reliability.
JJF
March 28th, 2008 at 5:29 pm
John,
Thank you for the kind words.
Do you think that this data can be analyzed the same way as the stock market? I’m thinking that there is a difference: this data is straight bean-counting, and the stock market is a matter of opinion - the collective opinion of investors, insiders, and traders.
Matt
March 31st, 2008 at 8:17 am
Hi Matt,
I’m charting the same factors as you, but using a one year smoothing on the data and then running a daily year over year transformation. I just saw your email address on another page so I’ll forward you my charts and my data from work and you can compare them against the data that you have. It’d be interesting to see if you notice any discrepancies in the data, or if it’s just the transformations on the data that are skewing our results.
Best regards,
Edward
March 31st, 2008 at 12:08 pm
Edward,
I experimented with the same smoothing that you have done: using the trailing 251 business days instead of a calendar year. I haven’t published that chart yet, but it looks different from my current calendar-year charts. It does end up in the same place though: a drop-off in 2008.
I am now of the opinion that both methods lack precision because of the monster tax-hauls that occur every-other Friday when most paychecks go out. So I am working on a new algorithm that will include an equal number of “monster days” and “regular days” in each data point.
Matt
April 3rd, 2008 at 10:33 am
Thanks for the data! If only those pdf’s were text files and we could see if the trends in withholdings match other recessionary periods.
April 3rd, 2008 at 12:02 pm
Mike,
I did, in fact, get the data as far back as it was consistently published. My code was able to parse the PDF’s OK, even though it wasn’t fun. There are a few more scattered reports prior to October 1998, but there are too many holes in the series so I didn’t use it.
It appears that when the reports were put online a few years ago, they couldn’t find them all, and put up what they had.
It is likely that the bean counters at the Treasury have ledger entries for this data going back much farther. There’s probably a big stack of moldy reports in a government warehouse somewhere.
Matt
May 1st, 2008 at 7:45 am
Hi Matt,
can you see or establish any correlation between the withholding taxes number and the initial claims and unemployment rate numbers?
Best regards,
Dressguard
May 1st, 2008 at 9:39 am
Hi Dressguard,
I haven’t studied that yet. However, it will be interesting to see how well they correlate since that might tell us something about how accurate the seasonal adjustments in the jobs numbers are.
Matt
May 2nd, 2008 at 10:57 am
Matt, your work here is great! Thanks for sharing it with everyone. One quick question for you: in the Q over Q chart above, is the right most bar (you know, the one that is down about 5.5%), is that bar measuring 2Q year-over-year data, or 1Q year-over-year data? Your footnote makes me think it’s 2Q data, but when I line up the bar with the date on the x-axis, it makes me think it might be 1Q data. Thanks for clarifying, and thanks for a great site. –Bill
May 2nd, 2008 at 11:10 am
Hi Bill,
Thanks for the kind words.
It is 2Q, and since 2Q is not over yet it is the first part of 2Q07 compared to the first part of 2Q08. The -5.5% will likely be bouncing around quite a bit more. See this page to zoom-in for more detail.
Matt
May 4th, 2008 at 1:22 am
Very thought provoking material Matt.
Does the series include receipts of estimated self-employment taxes?
If not, how can we adjust for the growing trend in outsourcing?
thanks again for your fine work.
May 4th, 2008 at 9:27 am
Hi Paul,
This series is not adjusted in any way. I will investigate the outsourcing issue in the future, however I don’t think there is a big rush. This data series seems to still be working quite well as it has turned down abruptly along with just about every other economic indicator back around January.
Matt
May 6th, 2008 at 4:30 am
Using data with this much noise requires applying moving averages.
May 6th, 2008 at 9:10 am
Hi Don,
That’s what I thought at first, but it turns out not to help. The moving-average charts that I made bounced around even more.
Matt
May 6th, 2008 at 1:44 pm
I am looking at a very sharp recession hitting soon, if it has not started already.
My elliott wave analysis on the sp 500 indicates that we could be in store for a 500 point selloff into the end of 2008.
Below is a link to a chart of how I think things might look in the years ahead.
http://www.visionproducts.us/Wave4%20SP50042708.jpg
May 7th, 2008 at 3:36 am
I am not sure how effective this indicator is.This is a lagging indicator because it’s based on employment.And employment always lags production which lags sales.Merely watching an indicator based on employment isn’t a great way to gauge the health of economy.The chart says 2002 there was negative growth in payroll taxes which is to be expected.But the economy infact had begun to turn around in Nov of 2001 and improved steadily in 2002.
May 7th, 2008 at 7:11 am
Hi Jagmohan,
I’m actually trying to predict the stock market rather than the economy. Perhaps the economy did turn in November 2001 as you say, but it wasn’t until a year later that the stock market turned up.
Perhaps leading indicators are not sufficient to break bear-market psychology. Maybe the market is only convinced after coincident and lagging indicators confirm the turn.
Matt