Friday, June 5, 2015

#NFPGuesses: A Recap and a year's summary

Swing and a miss! But this time it's a surprise in the opposite direction - I was 80,000 jobs too short.

Here's what the distribution of guesses looked like for this month's go-around:

One thing I've noticed is the remarkable consistency of the average guess. I've plotted out all of the rounds of #NFPGuesses from February to today, and notice how average has come out at around 225,000 each time.





As you can see, the average of guesses has been remarkably consistent, even when the actual numbers are not:

 It looks like the median twitter armchair economist is consistent, and consistently optimistic. In an average month, an average twitter user will guess +222K, which is higher than the actual jobs figure of 205,400.

There might be a few reasons why this might be the case:

  • This mid-winter dip coincides with severe weather in the Northeast, especially New England. It comes after a huge employment boom in autumn. So it is entirely possible, given a steady recovery and surprisingly disastrous conditions in a region, that tweeters would overestimate job growth.
  • Other job data appears to be strong, which means that NFP data might be underwhelming, relative to expectations. For example, JOLTS data (the Job Openings and Labor Turnover Survey), has been quite encouraging over the period I looked at. Companies are looking for employees harder than they have been in the past seven years.

  • Finally, this may simply be a manifestation of optimism bias, the tendency of people to predict good events to happen more frequently than reality, and bad ones less frequently. Finance, for example, has a natural incentive to be optimistic, given that good economic conditions prevail most of the time, and crises and recessions in fits and starts.
I want to keep my blog fresh, but I might have another post or 2 with insights I've gathered from looking at this particular trend. So stay tuned!

Until next time.

Thursday, June 4, 2015

#NFPGuesses Preview

Tomorrow will be the first Friday of the month, when the Bureau of Labor Statistics releases its employment numbers (called the Non Farm Payroll numbers). For twitter-addicted finance and economist types, this spawns a monthly ritual, #NFPGuesses, where people publicly post their guess for the month and see who can "nail the number".

For the past four months, I've been collecting every #NFPGuesses tweet in anticipation of collecting them and analyzing them. Over the past two weeks, I've learned an entirely new statistical package (Pandas for Python) and made a very simple, yet elegant and reusable tool to quickly extract and clean my tweets.

But let's get right into it. Here's a visualization of everyone's guesses from last month. I put myself in red:

Surprisingly, there are fewer extreme optimists/pessimists than I expected. This is likely because I filtered out accounts with fewer than 100 followers, but most of the outliers are either jokes:
or mismatched entries (human data entry would be more reliable, but python is faster!).

Twitter seems to be in-line with industry expectations, but when there's a miss, everybody misses. Last month, industry and twitter consensus was well above average– March was an awful hiring month!

It looks like Twitter is as good a forecaster as Wall Street. But this shouldn't surprise anyone, because macroeconomic forecasting is hard, NFP data is noisy (look at the revisions), and #NFPGuesses is "obscure" enough that only people who are already somewhat economically inclined will participate.

Stay tuned, because I'm going to do a lot more with this data... people should be turning in their guesses for tomorrow's numbers right now, and I've practically done all the heavy lifting already!

And oh, right...
See you soon!