Rookie Economist Errors: A Compendium

I begin instructing a pair lessons tomorrow; listed below are some cautionary notes.

From the first edition:

Forgetting whether or not nominal or actual magnitudes are extra vital

Usually we hear of data being damaged. However there are data and there are data. As an example, take into account this headline: Nasdaq and S&P 500 Climb to Records. The assertion is right, however it’s missing in context to the extent that the actual, or inflation adjusted, value is extra related (provided that share costs characterize valuations of a declare to capital).


Determine 1: Nasdaq (blue), and CPI-deflated (crimson), in 2009$. Supply: NASDAQ through FRED, BLS, and creator’s calculations.

Now, some people proof wariness about deflation, notably by way of the accuracy of the deflators. Nonetheless, it’s normally higher to account for value stage results and be aware the issues, fairly than depend on individuals to maintain their most popular deflators of their head to calculate the actual magnitudes (see this post).

Including up chain weighted portions

Whereas in lots of instances, actual magnitudes are the related ones, working with actual magnitudes isn’t at all times simple. As an example, in precept, if one had knowledge on actual consumption by households in Wisconsin and family in Minnesota, and one wished so as to add them as much as discover actual consumption in Wisconsin and Minnesota, one might try this by deflating every state’s consumption by the CPI and including. That’s true as a result of the CPI is actually a set base 12 months weighted index (on this case, a Laspeyres index, utilizing the preliminary weights).

This isn’t true for the collection within the nationwide earnings and product accounts, such because the parts of GDP. The true measures – consumption, funding, authorities, internet exports — are obtained utilizing chain weighted deflators, i.e., deflators the place the weights range over time.

Can one make a mistake this fashion? Definitely. Think about this recent case at Political Calculations, whereby the commentator added up the state stage actual GDPs and since it didn’t match the nationwide stage GDP, the commentator inferred an enormous impending downward revision in GDP.


Supply: Political Calculations.

Evidently, the huge downward revision didn’t happen. Total, actual GDP was revised barely upward. Determine 2 depicts what truly transpired.


Determine 2: Actual GDP pre-benchmark revision (inexperienced), post-benchmark revision (crimson), and arithmetic sum of state stage GDP (black), all in Ch.2009$ SAAR. Supply: BEA 2016Q1 third launch, 2016Q2 advance launch, BEA state stage quarterly GDP, revision of 27 July 2016, and creator’s calculations.

Extra dialogue in this post. Now, generally one desires to precise ratios of actual magnitudes. This post discusses a method at getting at one thing like that.

Forgetting what “SAAR” means

SAAR is brief for Seasonally Adjusted, at Annual Rates. Most US authorities statistics are reported utilizing this conference, even when the information are at a month-to-month or quarterly frequency. (In distinction, European quarter-on-quarter GDP figures are sometimes reported on a non-annualized foundation.) Most prominently, quarterly GDP is reported at annual charges, so when one sees 18,000 Ch.09$ in 2016Q2, that doesn’t imply the movement of GDP was 18,000 Ch.09$ in that quarter; fairly that if the movement that occurred in 2016Q2 continued for an entire 12 months, then GDP could be recorded at 18,000 Ch.09$.

Now, this doesn’t matter if one is calculating share progress charges (so long as one additionally remembers to annualize the expansion charge if one is calculating quarter-on-quarter modifications). It does matter if one is calculating a “multiplier”, the rise in GDP for a given improve in authorities spending. That’s as a result of the federal government spending improve (or stimulus) is usually reported in absolute (non-annualized) charges, and GDP in SAAR phrases. Clearly, if one did the mathematical calculation forgetting this level, one’s multipliers would look 4 occasions as giant as they need to. If compounded with failing to have in mind annualizing progress charges, then they might look sixteen occasions as giant as they need to. Even skilled economists make this error – take into account the case of this University of Chicago economist, who thought “…the multiplier is 20 or 50 or one thing like that” as a result of he was primarily dividing a quarterly stimulus determine by an annualized determine, and forgetting that progress charges are sometimes reported at annualized charges.

What in regards to the “SA” a part of the SAAR? More often than not, one desires to make use of seasonally adjusted collection; in reality, that is nearly at all times what’s reported within the newspapers. The reason being that there’s a huge seasonal element to many financial variables; retail gross sales bounce in December due to the Christmas holidays, for instance.

Often, individuals (normally noneconomists) get into hassle after they combine and match seasonal and non-seasonal knowledge. As an example, Wisconsin Governor Walker’s campaign got into trouble after they touted job creation numbers obtained by adding together seasonally unadjusted jobs figures (from what is known as the Quarterly Census of Employment and Wages) with seasonally adjusted jobs figures (from the institution survey) to get a cumulative change in employment. (They did this as a result of QCEW figures lag by many months, whereas the institution survey knowledge are extra well timed). That is proven in Determine 3.


Determine 3: Wisconsin nonfarm non-public employment from Quarterly Census of Employment and Wages, not seasonally adjusted (blue), non-public nonfarm non-public employment from institution survey, seasonally adjusted (crimson). Black arrows denote modifications over QCEW and institution survey figures; teal arrows over institution survey. Supply: BLS.

Discover that one can calculate the modifications from December 2010 (simply earlier than Walker takes workplace) to March 2012 (the newest QCEW figures out there as of December 12, 2012), after which add to the change from March 2012 to October 2012 (the newest institution determine out there as of December 12, 2012). That’s, add 89.1 to six.4 to get 95.5 thousand, near the 100 thousand determine cited by Governor Walker’s marketing campaign. You may see why Governor Walker’s marketing campaign officers did so – the right calculation utilizing the change within the institution survey from December 2010 to October 2012 was solely 61.1 thousand.

Log aversion

Oftentimes, we depict financial statistics in log phrases. The explanation for that is that when plotted over time, a variable rising at a continuing charge will seem like it’s accelerating if the Y-axis is expressed in stage phrases. Nonetheless, it would seem like a line with fixed slope if the Y-axis is in log phrases.


Determine 4: Actual consumption, normalized to 1967Q1=1 (blue), and log actual consumption, normalized to 1967Q1=0 (crimson). Supply: BEA, 2016Q2 second launch, and creator’s calculations.

Solely inspecting the extent collection, consumption seems to be accelerating, notably within the 2000’s. That illusory acceleration disappears within the log collection.

For extra on log transformations, see [1][2][3][4], and [5]. And right here’s Jim Hamilton’s post on logs. Right here is an instance of the place using logs has pushed somebody to ranting and raving.

Shadowstats and different knowledge conspiracies

It isn’t unusual for commentators to allege conspiracies to govern authorities financial statistics. Take this FoxNews article:

What a coincidence. Simply as momentum was constructing in direction of an rate of interest hike by the Fed, alongside comes a dismal jobs report that takes any improve off the desk. Opposite to the final notion, it is a fortunate break for Democrats. … Given all that’s stake, it’s stunning that nobody has questioned whether or not the roles report may need been massaged by the Labor Division.

In fashionable occasions, these kind of allegations are unfounded. The info collection may not be notably correct, however deliberate manipulation to distort the financial image doesn’t happen for traditional collection launched by the BEA, BLS, and Census.

One notably egregious type of conspiracy-mongering is Shadowstats, a money-making enterprise that purports to offer a extra correct set of value measurement. As a substitute of going into element, I’ll flip the case over to Jim Hamilton, who thoroughly debunks the Shadowstats approach. Anyone who cites Shadowstats ought to instantly lose all credibility. So … don’t do it!

Extra knowledge conspiracies, see here.

From the second edition:

Don’t Purpose from identities

Among the best recognized identities in economics is the definition of GDP:

Y ≡ C + I + G + X – IM

From this, a writer at the Heritage Foundation deduces the next:

Congress can’t create new buying energy out of skinny air. If it funds new spending with taxes, it’s merely redistributing current buying energy (whereas reducing incentives to provide earnings and output). If Congress as a substitute borrows the cash from home traders, these traders can have that a lot much less to speculate or to spend within the non-public financial system. In the event that they borrow the cash from foreigners, the steadiness of funds will modify by equally elevating internet imports, leaving whole demand and output unchanged. Each greenback Congress spends should first come from elsewhere.

In different phrases, G rising by one greenback essentially reduces both C or I by one greenback. In fact, that is true holding Y fastened. There isn’t any cause why this could essentially be true. One can’t say what occurs and not using a mannequin.

The creator provides in one other id, the steadiness of funds, for good measure:

BP ≡ CA + FA + ORT

The place CA is the present account (roughly the commerce steadiness, X-IM), FA the non-public monetary account, and ORT official reserve transactions. FA going up by one greenback outcomes essentially in CA declining by one greenback within the creator’s most popular interpretation. In fact, his holds ORT fixed (no modifications in international change reserves). And it guidelines out repercussion results, such that offsetting lending happens…

In different phrases, there isn’t any option to keep away from utilizing some kind of mannequin when one desires to impute trigger and impact. It doesn’t need to be mannequin with equations, but when one tries to keep away from utilizing a mannequin, one finally ends up implicitly utilizing a mannequin, that as a rule, has inner inconsistencies, or implausible assumptions.

Don’t Neglect to Test for Information Breaks

On this period of simply downloadable knowledge, the analyst is tempted to skip to the calculations earlier than studying the footnotes. That is problematic, as a result of, as authorities and worldwide statistical businesses gather knowledge, the mode of the information assortment or the technique of calculation generally change. These modifications are normally famous, but when one doesn’t learn the documentation, one could make severe errors. As an example, inspecting civilian employment (FRED collection CE16OV), one would suppose there was an amazing increase in employment in January 2000.


Determine 1: Civilian employment (blue, left scale) and alter in civilian employment (crimson, proper scale), each in 000’s, seasonally adjusted. Supply: FRED collection CE16OV.

If one inspects different collection, say nonfarm payroll employment, one sees no corresponding bounce. This implies the increase interpretation is flawed. Additional proof of a “break” is discovered by inspecting the primary distinction of the collection (in crimson). The spike in January 2000 is a 1.5% change (m/m), whereas the the usual deviation of modifications is 0.3% (calculations in log phrases).

In truth, the bounce is as a result of introduction of latest inhabitants controls related to the Census. New controls are utilized each decade, so it is a recurring (and recognized — to those that are cautious) downside. Nonetheless, right here’s an example of the mistake regarding the participation figure. Different breaks are much less apparent. It is a cautionary be aware to all who obtain knowledge with out consulting the documentation.

Don’t Make Absolute Predictions When Revisions Abound

Think about Professor Casey Mulligan on October 26, 2008:

In accordance with the BLS, nationwide nonfarm employment was 136,783,000 (SA) on the finish of 2006, because the housing value crash was getting underway. Actual GDP was $11.4 trillion (chained 2000 $). Barring a nuclear battle or different violent nationwide catastrophe, employment won’t drop beneath 134,000,000 and actual GDP won’t drop beneath $11 trillion. The numerous economists who predict a extreme recession clearly disagree with me, as a result of 134 million is simply 2.4% beneath September’s employment and solely 2.0% beneath employment in the course of the housing crash. Time will inform.

Time has instructed. Here’s what truly occurred.


Determine 2: Nonfarm payroll employment, September 2008 launch (blue), and September 2016 launch (black), and 134,000,000 employment flooring cited by Casey Mulligan (crimson). NBER outlined recession dates shaded grey. Supply: BLS through FRED and ALFRED.

Right here is the place figuring out about revisions is vital. Not solely was lastly revised employment 537,000 beneath what was estimated as of end-October; employment was falling a lot quicker than estimated on the time. For the three months ending in August 2008, employment was falling 309,000/month, fairly than 99,700/month.

Occupied with the information as settled numbers, fairly than estimates, can result in embarrassingly inaccurate conclusions, to be lengthy immortalized on the web.

Don’t Do Easy Subtraction of Chain Weighted Measures

At Political Calculations, Ironman has focused on drought and manufacturing (particularly aircraft) as a proof for Kansas’s lagging financial fortunes. Particularly, he asserts that actual Kansas GDP stripped of agriculture and manufacturing appears to be like significantly better. Sadly, his graph in his post plots a collection the place he calculated Kansas GDP ex-agriculture and manufacturing by merely subtracting actual agriculture and actual manufacturing — each measured in Chain weighted {dollars} — from actual GDP measured in Chain weighted {dollars} (the crimson line in Determine 3 beneath). That is, fairly plainly, the flawed process, as I defined in this post.


Determine 3: Log Kansas actual GDP ex. agriculture and manufacturing, calculated utilizing Törnqvist approximation (blue), and calculated utilizing easy subtraction (crimson), 2011Q1=0. Dashed line at 2011Q1, Brownback takes workplace. Mild tan shading denotes interval throughout which Ironman identifies as drought. Supply: BEA and creator’s calculations.

Utilizing sturdy manufacturing as a substitute of whole manufacturing doesn’t change the outcomes. In different phrases, Ironman’s conclusion is totally overturned when utilizing the proper calculation. So, beware making conclusions while you don’t perceive the information!

Don’t Defend Factoids as Info

On this period of the web, it’s vital to keep in mind that the boundaries to circulating misinformation are very low. Right here is Governor Romney making a factual assertion, as quoted in Salon (see additionally WSJ):.

“We needs to be seeing numbers within the 500,000 jobs created per thirty days. That is method, method, method off from what ought to occur in a traditional restoration.

That is patently, flawed, as mentioned in this post. However right here is frequent commenter Rick Stryker making an attempt to alter the phrases of debate:

Counting the variety of occasions that month-to-month employment will increase have been larger than 500K since 1939 is simply an try and decrease the bar for what administration supporters know of their hearts is a failed presidency.
In 1939, the dimensions of the labor pressure was 30 million; right this moment it’s 133 million. A 500K month-to-month improve on 30 milion [sic] could be actually gigantic and by no means what Romney was advocating. To place Romney’s comment in context, we have to modify for the present measurement of the labor market, which we will merely do by dividing 500K into 133 million, yielding 0.38%. Going again to 1939, there are 172 instances wherein month-to-month employment features have been not less than 0.38% of the present labor market measurement. Admittedly, the final time we noticed sustained will increase of 0.38% of present labor market ranges was in the course of the restoration from the 1982 recession. However that simply implies that Romney is setting an bold however not traditionally unreasonable employment objective. Given the dimensions of the present employment gap, it is usually obligatory objective if we’re to return to full employment anytime quickly.

Word how Mr. Stryker tries to re-state Governor Romney’s assertion to make it appear extra affordable, in proportional phrases. As I reveal in this post, the five hundred,000 quantity nonetheless stays clearly flawed, even after adjusting for labor pressure measurement. That is proven within the determine I generated on the time:


Determine 3: Log nonfarm payroll employment relative to 2009M06 trough (blue), to 2001M11 trough (crimson), to 1991M03 trough (inexperienced), 1982M10 trough (orange). Lengthy dashed line at 2012M05 on the time of Governor Romney’s remarks. Supply: BLS, Might 2015 launch, and creator’s calculations.

Then the declare turns into aspirational, fairly than factual. Quoting Mr. Stryker again:

Setting bold however affordable targets is simply what a future President needs to be doing. Supporters of the present administration perceive that they should decrease expectations nonetheless. It was amusing to learn erstwhile ecstatic now lukewarm Robert Deniro’s remarks to NBC’s David Gregory, which in all probability sums it up effectively for administration defenders: “’s not simple to be President of the US.” And: “I do know he’ll do higher within the subsequent 4 years…”

This course of of fixing the goalposts is the surest signal of somebody figuring out they’ve misplaced the argument, however refusing to confess error. Even Governor Romney subsequently modified his declare to 250,000 new jobs/month as being customary.

So children, don’t be a “Rick Stryker”. Admit while you’ve made a mistake.

For different errors to keep away from, see Rookie Economist Errors

Replace, 10/13 4:45pm Pacific: The pseudonymous Rick Stryker retains on digging. Under are two graphs — the primary of month-on-month employment progress, and the 0.38% threshold that Stryker mentions as equal to Governor Romney’s 500,000/month employment progress, and the second as trailing 12 month-to-month common. The second highlights simply how uncommon it’s for the five hundred,000/mo determine to be realized over something like a steady timeframe in current many years, even when expressing in proportional phrases as Rick Stryker suggests.


Determine 4: Month-on-month progress charge of nonfarm payroll employment (blue), and 0.38% threshold (crimson line). NBER outlined recession dates shaded grey. Supply: BLS, NBER, and creator’s calculations.


Determine 5: Trailing twelve month month-to-month common progress charge of nonfarm payroll employment (blue), and 0.38% threshold (crimson line). NBER outlined recession dates shaded grey. Supply: BLS, NBER, and creator’s calculations.
Appears like one has to go method again to get the five hundred,000 norm to be affordable…

Replace, 10/17 4:15pm Pacific: Within the curiosity of comprehensiveness, let me be aware an error I made, involving Rounding Errors (first famous in this post).

In precept, actual amount = whole worth/value deflator. As an example GDP09 = GDP/PGDP09, the place GDP is measured in $, PGDP09 is the GDP deflator taking up a price of 1 in 2009, and GDP09 is GDP measured on 2009$. In apply, there’s a slight rounding error, which generally doesn’t make a distinction, however can if (1) progress charges are very excessive (or very destructive), and (2) one is annualizing quarterly progress charges.

I used the manually deflated collection for the 2008Q4 q/q calculation, when on this case it might have been higher to make use of the actual collection reported by BEA to do the calculation.

Figures 6 and seven present the rounding errors.


Determine 6: GDP in bn. Ch.2009$ SAAR, calculated by deflating nominal GDP with the GDP deflator (blue) and as reported by BEA (crimson), FRED collection GDPC1. Supply: BEA 2014Q2 2nd launch and creator’s calculations.


Determine 7: Quarter-on-quarter annualized progress charge in GDP in bn. Ch.2009$ SAAR, calculated by deflating nominal GDP with the GDP deflator (blue) and as reported by BEA (crimson), FRED collection GDPC1. Supply: BEA 2014Q2 2nd launch and creator’s calculations.

But extra errors:

Don’t dismiss seasonality.

Don’t be casual about estimated trends.

Don’t make policy analysis based on not-statistically-significant parameters.

Don’t forget that just because somebody calls something a “competitiveness index” doesn’t mean it actually measures competitiveness.

Do not attribute all things you don’t believe in to “conspiracy”, before checking the data yourself. (By the way in which, simply because there’s no conspiracy doesn’t imply the information are unbiased or noisy).

Leave a Reply

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