Showing posts with label US data. Show all posts
Showing posts with label US data. Show all posts

Saturday, April 11, 2020

11/4/20: Covid19 vs Seasonal Deaths: End of an Argument


A large number of social media pontificators have been making various claims that COVID-19 pandemic is in some ways related to the seasonal flu. One of the arguments made in this context is that the pandemic-related deaths are similar to those produced by a seasonal flu and therefore do not warrant drastic measures deployed by the governments around the world.

Setting this argument to rest, the NY Times have published this analysis: https://twitter.com/sangerkatz/status/1248685404775239680?s=20 with an associated chart:



Others make an argument that people dying from COVID-19 are somehow the same people who would have died from a different condition anyway, without the virus pandemic.

Given the levels of utilization of hospital beds in countries like France, Italy, the UK, and the U.S., and in fact across the entire range of advanced economies, it is hard to imagine how anything about the COVID-19 pandemic can be considered at even remotely related to 'normal' or 'seasonal'.

Monday, September 19, 2016

19/9/16: US Median Income Statistics: Losing One's Head in Cheerful Releases


In our Business Statistics MBAG 8541A course, we have been discussing one of the key aspects of descriptive statistics reporting encountered in media, business and official releases: the role that multiple statistics reported on the same subject can have in driving false or misleading interpretation of the underlying environment.

While publishing various metrics for similar / adjoining data is generally fine (in fact, it is a better practice for statistical releases), it is down to the media and analysts to choose which numbers are better suited to describe specific phenomena.

In a recent case, reporting of a range of metrics for U.S. median incomes for 2015 has produced quite a confusion.

Here are some links that explain:


So, as noted on many occasions in our class: if you torture data long enough, it will confess... but as with all forced confessions, the answer you get will bear no robust connection to reality...

Thursday, May 12, 2016

12/5/16: Leaky Buckets of U.S. Data


Recently, ECB researchers published an interesting working paper (ECB Working Paper 1901, May 2016). Looking at the U.S. data that is released under the embargo, they found a disturbing regularity: across a range of data, there is a strong evidence of a statistical drift some 30 minutes prior to the official time of the release. In simple terms, someone is getting data ahead of the markets and is trading on it in sufficient volumes to move the market.

Let’s put this into a perspective: there is a scheduled release for private data that is material for pricing the market. The release time is t=0. Some 30 minutes before the official release, markets start pricing assets in line with information contained in the data yet to be released. This process continues for 30 minutes until the release becomes public. And it moves prices in the direction that correctly anticipates the data release. The effect is so large, by the time t=0 hits and data is made publicly available, some 50% of the total price adjustment consistent with the data is already priced into the market.

"Seven out of 21 market-moving announcements show evidence of substantial informed trading before the official release time. The pre-announcement price drift accounts on average for about half of the total price adjustment,” according to the research note.

Pricing occurs in S&P and U.S. Treasury-note futures and data sample used in the study covers January 2008 through March 2014.

Here is the data list which appears to be leaked in advance to some market participants:

  1. ISM non-manufacturing
  2. Pending home salses
  3. ISM manufacturing
  4. Existing home sales
  5. Consumer confidence from the Conference Board (actually, CB has taken some actions recently to tighten their releases policy)
  6. Industrial production (U.S. Fed report)
  7. The second reading on GDP
  8. There is also partial evidence of leaks in other data, such as retail sales, consumer price inflation, advance GDP estimates and initial jobless claims. 

Overall, plenty of the above data are being released by non-private (aka state) agencies.

The authors control for market expectation, including forecasts drift (as date of release grows nearer, forecasts should improve in their accuracy, and this can have an effect on market pricing). They found that “more up-to-date forecasts” are no “better predictors of the surprise” than older forecasts. In addition, as noted by the authors: “these results are robust to controlling for, among others, outliers, data snooping, nearby announcements and the choice of the event window length.”

The problem is big and has gotten worse since 2008. “Extending the sample period back to 2003 with minute-by-minute data reveals both a higher announcement impact and a stronger pre-announcement drift since 2008, especially in the S&P E-mini futures market. Based on a back-of-the-envelope calculation, we estimate that since 2008 in the S&P E-mini futures market alone the profits associated with trading prior to the o fficial announcement release time have amounted to about 20 million USD per year.”

Two tables summarising there results.




The paper is available here: http://www.ecb.europa.eu/pub/pdf/scpwps/ecbwp1901.en.pdf?ca0947cb7c6358aed9180ca2976160bf