forex-scalping.com

US weekly jobless claims are seasonally adjusted, but holiday weeks often produce distorted figures.

The 4-week average is widely used because it smooths single-week distortions. Continuing claims, released at the same time, show how many people stay on benefits and can confirm whether a trend in initial claims is real.

Around Thanksgiving, Christmas and New Year, the release day itself can also shift, for example to Wednesday instead of Thursday. Checking the calendar for the actual release time in holiday weeks avoids being caught by an unexpected print.

Why:

• Fewer working days affect filing patterns.
• Seasonal adjustment factors may not match current patterns.
• State offices may close.

Implications:

• Large jumps or drops in holiday weeks are often reversed the following week.
• The 4-week average helps smooth distortions.
• Market reactions to holiday-week claims are often smaller.

Do you adjust your reading of claims during holiday weeks?

In holiday weeks I stop reading the adjusted number against consensus and compare the unadjusted figure with the same week a year earlier. If unadjusted claims are within a few thousand of last year's, a 25k headline jump is mostly the seasonal factor talking.

The same trick helps in early July. Car plants retool around then and the timing shifts from year to year, so the adjusted series can throw a spike unrelated to the labour market.

Continuing claims cover the week before the initial claims week, which is easy to forget when lining the two up. In a holiday stretch one of them can fall in the distorted week while the other does not.

Thanksgiving week normally moves the release to Wednesday, which this year should be 25 November. That day is usually crowded with other US data and thin pre holiday liquidity, so I skip claims entirely.

Reactions are smaller, as the post says, but a clean reversal the following week is not guaranteed. Holiday distortions can take two or three weeks to wash out, especially around the New Year.

Does anyone track the state level figures to catch one state driving the number? I have only just started on that table and it is slow going.