Reform Retro!

A key concept in Statistical Analysis and the Scientific method is the “Null Hypothesis.” This is the belief that whatever happened in the past will continue to happen in the future. The larger the sample size is, the more results tend to regress towards the mean. Thus, with big samples, both scientists and insurance specialists can make accurate predictions about future events based in statistical analysis of past events. Given that Retro groups and non-Retro groups involve huge numbers of nearly identical workers, it is scientifically and statistically unlikely that Retro groups would out-perform non-retro groups.

 

This is why I have a hard time believing that the “experts” at L & I (who presumably understand statistical analysis) could have permitted a “1994 computer coding error” to fool them into believing that the Retro program was performing some kind of miracle which would permit a billion dollars in undeserved Retro Refunds over a 15 year period of time.

 

Nevertheless, L & I developed a complex scheme which permitted these un-deserved “gifts” from the State tax payers by using a “Performance Adjustment Factor” (or PAF) to make “adjustments” to the normal equation of Refunds = Premiums minus Developed Claims.

 

The following is L & I’s description of this (absurd) system (slightly edited for clarity) and taken from the L & I website:

 

Developed Losses for every Retro claim are influenced by a "performance adjustment factor" (PAF) which L&I calculates for each coverage period. The PAF is a multiplier which adjusts the developed losses for Retro claims, so that the aggregate refund (net of total refunds and assessments) is proportional to how well all the Retro employers out-performed all the non-retro employers.

If the Retro participants on the whole out-performed everyone else by $50 million, the PAF (and, therefore, developed losses) are set at the exact level they need to be so that L&I makes a $50 million net refund to the Retro employers. ..Aggregate refunds have been determined solely by retro vs. non-retro performance. PAF's are used by themselves as multipliers to determine developed losses. If the PAF is .9, then the developed losses for a $400,000 total permanent disability claim would be only $360,000.

What L & I was saying was that even a 10% difference between a Retro and non-retro program can result in an increased refund in the above example of $40,000. But for the L & I formula to work, two conditions must be met:

 

First, L & I must accurately determine the cost of non-Retro programs (which it failed to do because of the “computer coding” error and other problems described later).

 

And second, there must actually be a difference between Retro and non-Retro programs (which as noted above is unlikely and as described below has been proven to not exist).

Here is another statement from L & I which should have been a clue something was wrong:

Performance adjustment factors have always been less than 1.00. Theoretically, they could be greater than 1.00, but never have been. This means that developed losses have been lower and refunds have always been higher than they would have been without this process.

 

In other words, using the crazy PAF formula described above, Retro Programs have always out-performed non-Retro programs every year during the past 15 years. This is like playing cards where one person always beats the other for 15 straight years.

 

Statistically, this is unlikely and instead is a clue that the deck has been stacked! Either that or the dealer is in on the take. In the present case, the computer coding error was one way of stacking the deck. But even this would not have worked without the dealer (L & I) being in on the take.

 

Put in plain English, the Department of Labor and Industries claims they are using Performance adjustment factor (PAF) to balance the “Loss Ratio” (or Claims divided by Premiums) of Retro programs compared to the Loss Ratio of Non-Retro programs.

 

But is this really what happens?

 

The following chart compares the Retro Loss Ratios and Refund Ratios for the Third Quarters for the Years 2002 to 2007 (note that BIAW reports in the 3^rd Quarter. Also the other three quarters are small and thus are unstable):

 

Year 3rd Q only	Retro Premiums ($M)	Retro Claims ($M)	Prem - Claims	Retro Loss Ratio %	Retro Refund Ratio %	PAF	Retro Refunds ($M)	Cost to Taxpayers ($M)
2002	315	299	+16	95	23	0.82	73	57
2003	400	364	+36	91	25	0.99	102	66
2004	480	460	+20	96	22	1.09	108	88
2005	532	497	+35	94	24	1.13	130	95
2006	567	557	+10	98	20	1.06	112	102
2007	558	592	-34	106	14	1.08	76	110

 

There are several unusual things about the above chart.

 

First note that PAF’s have been above 1.000 ever since 2004. Thus the quote from the L & I website about PAF’s always being less than 1.000 is not correct and thus this section of the L & I website has not been updated since before 2003.

 

Second, regardless of whether the PAF is below one or above one, the Retro Refunds keep skyrocketing. Thus, the tax payer subsidies do not appear to be related either to the Retro loss ratios or the Performance Adjustment Factors.

 

Third, the cost to the tax payers (which is the cost of the refunds minus the premiums to claims difference) also keeps skyrocketing. Even in 2007, when the claims exceeded the premiums by $34 million dollars, there was still a Retro “Refund” of $76 million dollars for a total cost to tax payers of $110 million dollars. How can this be justified?

 

L & I claims it is because Retro programs are ‘saving money”. But if this were true, then why is the Retro cost to Non-Retro cost ratio (as shown in the PAF) greater than one during the years 2004 through 2007?

 

If Retro were really saving money, then the PAF would be less than one.