Read on Twitter
. @BLS_gov changed seasonal adjustment and has not provided any conversion factors. We do.
Bottom line: new procedure lowers estimated new UI claims
old method would have given 1010K
new method given 881K
https://abs.twimg.com/emoji/v2/... draggable="false" alt="🧵" title="Thread" aria-label="Emoji: Thread">: why seasonally adjust at all, old method, new method, why change
Bottom line: new procedure lowers estimated new UI claims
old method would have given 1010K
new method given 881K
: why seasonally adjust at all, old method, new method, why change" title=". @BLS_gov changed seasonal adjustment and has not provided any conversion factors. We do.Bottom line: new procedure lowers estimated new UI claimsold method would have given 1010K new method given 881K https://abs.twimg.com/emoji/v2/... draggable="false" alt="🧵" title="Thread" aria-label="Emoji: Thread">: why seasonally adjust at all, old method, new method, why change" class="img-responsive" style="max-width:100%;"/>
New claims for unemployment insurance are one of the only *weekly* indicators we have of the health of the economy. Also the most timely. They were first economic series to spike when covid hit.
But claims also have a seasonal pattern. They rise in Dec/Jan as workers leave temporary holiday jobs. They also rise in the summer. These surges don’t indicate broader economic distress. So we want to adjust for them.
Take last year
Sep 7 2019: 160K
Jan 11 2020: 339K
Yet the economy was very healthy! So we want to “smooth” out the seasonality.
Enter “seasonal adjustment”. Two methods:
1) multiplicative: 339/160 = 2.12
2) additive: 339 - 160 = 179
Sep 7 2019: 160K
Jan 11 2020: 339K
Yet the economy was very healthy! So we want to “smooth” out the seasonality.
Enter “seasonal adjustment”. Two methods:
1) multiplicative: 339/160 = 2.12
2) additive: 339 - 160 = 179
The BLS has been doing multiplicative (method 1), but this week switched to additive (method 2). Multiplicative was a better model for handling the secular downward trend in UI claims from 2011-2019.
But multiplicative generates wacky patterns now.
Take this week, where we had 833K claims not seasonally adjusted. If we continue the toy example from above
multiplicative: 833K*2.12 = 1766K
additive: 833K + 179K = 1012K
Very different! (remember this is just a toy example)
Take this week, where we had 833K claims not seasonally adjusted. If we continue the toy example from above
multiplicative: 833K*2.12 = 1766K
additive: 833K + 179K = 1012K
Very different! (remember this is just a toy example)
Which adjustment is better? With the unprecedentedly large shock, the multiplicative adjustment dramatically over-corrects by amplifying the seasonality. We don& #39;t think that seasonality 5 times larger during covid. So multiplicative doesn& #39;t make sense now (it did pre-crisis).
The natural thing would be for @BLS_gov to provide both additive and multiplicative going backward. But they aren& #39;t and I& #39;m unclear on why not. This is a problem because we want to compare the *change* in UI claims and BLS isn& #39;t enabling us to do this.
We don& #39;t have BLS& #39; exact seasonal algorithm, but we can come extremely close for multiplicative using the STL algorithm. So we have produced both series for 2020.
https://docs.google.com/spreadsheets/d/1dQP56ROCUIHq6ND1EPLWbCp6XITQ8RafY4kFPFz-6NU/edit?usp=sharing">https://docs.google.com/spreadshe...
https://docs.google.com/spreadsheets/d/1dQP56ROCUIHq6ND1EPLWbCp6XITQ8RafY4kFPFz-6NU/edit?usp=sharing">https://docs.google.com/spreadshe...
Our additive model is very close too.
Not seasonally adjusted: 833K
Us additive: 876K
BLS additive: 881K
h/t to our RA @pd_robertson who did these calculations
Not seasonally adjusted: 833K
Us additive: 876K
BLS additive: 881K
h/t to our RA @pd_robertson who did these calculations
Additive produces smaller adjustments than multiplicative and this is a good thing.
An open question is whether we should be using seasonal adjustment factors at all right now. @nytimes has decided to ignore these factors (per @bencasselman) https://twitter.com/bencasselman/status/1300845265247711233">https://twitter.com/bencassel...
An open question is whether we should be using seasonal adjustment factors at all right now. @nytimes has decided to ignore these factors (per @bencasselman) https://twitter.com/bencasselman/status/1300845265247711233">https://twitter.com/bencassel...
I think this might be a mistake. Claims surge from September to December even in normal economic times. NYT might report "bad news, economy is deteriorating" but in fact the increase in claims could reflect seasonal patterns.
On the other hand, rising claims at the end of the year could be bad news! Maybe workers aren& #39;t even getting hired for seasonal jobs at all and so there will be no seasonal increase in claims this fall. In the future, we can use industry composition of UI claims to assess this.
If there& #39;s substantial interest, we might decide to update both multiplicative and additive seasonal adjustment in future weeks.
FYI @FrancoisGourio has a nice piece on seasonality that we found helpful https://www.chicagofed.org/publications/economic-perspectives/201">https://www.chicagofed.org/publicati...
FYI @FrancoisGourio has a nice piece on seasonality that we found helpful https://www.chicagofed.org/publications/economic-perspectives/201">https://www.chicagofed.org/publicati...
