Random Fluctuations

MathJax

2012-01-19T20:44:00.000-05:00

In the latest issue of the Notices of the American Mathematical Society, there is an article on MathJax, a project to help nicely render mathematics on the web, using TeX or MathML. I decided to see if I can get it working here on blogspot, so here goes nuthin': $$\LARGE TVaR_\alpha =\frac{\int_{F^{-1}(\alpha)}^\infty x\;f(x)\;dx}{1-\alpha}$$ NICE! I'm using the MathJax content delivery network; well, actually, the secure connection (of course). Once that is done, then $$...$$ or \[...\] delimit displayed mathematics, and $...$ delimits in-line math. What should also work is right-clicking the image to see the source, although that may only function if MathJax is installed locally. Regardless, this should make it easier to display mathematics on the web, and should obviate the need to have images rendered as gifs and loaded (as can be seen in previous posts. Maybe I'll post a bit more often now!

Bayesian analysis: Comparing algorithms Part 1?

2011-08-22T18:46:00.001-04:00

I recently had the opportunity to engage in some Bayesian analysis at work. I was able to state the problem in terms of the lognormal distribution, and took advantage of JAGS and its integration with "R" using the R2jags package. The client was very happy, and life moved on. However, as the underlying problem may be better described by a transformed beta than a lognormal undergoing transforms (say that five times quickly!), it got me to thinking about how to teach myself at least rudimentary Bayesian analysis using MCMC methods.As my main text, I have been using Dr. Scott M. Lynch's excellent text, "Introduction to Applied Bayesian Statistics and Estimation for Social Scientist". Starting small (I haven't worked on the four-parameter distribution yet), I decided to play with data for which I knew the answer, and see if I could write my own Gibbs or Metropolis-Hastings algorithms. I picked a gamma distribution as my test case.

I'm no expert on MCMC methods, so I'm not going to even attempt to explain how this works, but I did see how the different algorithms behave, and it is rather insightful. For the gamma distribution, while I always parameterize it using shape and scale, for the purposes of Gibbs sampling, I used shape/rate. The nice thging about Gibbs sampling is that as long as your can sample from a distribution which is proportional to the actual conditional distribution, you're fine, and do not have to calculate the constants of proportionality that bring the distribution back to an integral of 1.

In the case of the gamma distribution, and using a flat prior, it is the likelihood which dominates the conditional distribution. Using the shape/rate parameterization, , the likelihood over the observations is:

When looking for a conditional distribution for alpha, this does not resolve into anything remotely nice (although see Fink (1997) "A Compendium of Conjugate Priors" (pdf) for more) so Gibbs sampling will have to give way to a Metropolis algorithm. I'm not sophisticated enough to use a non-symmetrical proposal distribution, so it is not a Metropolis-Hastings algorithm, maybe in a future post. However, when looking at beta, and discarding the x's and alphas, the core of the likelihood is rather gamma-like. Actually, it is proportional to a gamma distribution with the rate equal to the sum of the observed data points and the shape equal to one more than the product of the number of observations and the underlying alpha.

With that observation, and properly writing the algorithm (which took more time than I care to admit), a comparison between using Gibbs sampling or straight Metropolis on the scale parameter can be made.

Here is a "pure" Metropolis algorithm to calculate the posterior distribution, assuming the input data is gamma (this and all "pretty" R code created by Pretty R at inside-R.org :

MHGam <- function (x, n.iter=10000, n.burnin=0){
alpha <- matrix(.5, n.iter)
theta <- matrix(10, n.iter)
acca<- 0
accq <- 0
lnpost <- function(a, q, z){
  NLL <- (a-1)*log(z)-(z/q)-a*log(q)-lgamma(a)
  return(sum(NLL))
}
for (i in 2:n.iter)
{
#sample theta
  acc <- 1
  theta[i] <- theta[i-1] + rnorm(1, mean=0, sd=100)
  if (theta[i]<=0) {acc=0; theta[i] <- theta[i-1]}
  LL <- lnpost(alpha[i-1], theta[i], x)
  if ((LL - lnpost(alpha[i-1], theta[i-1], x)) < log(runif(1))) {
    theta[i] <- theta[i-1]
    acc <- 0
  }
  accq <- accq + acc
#sample alpha
    acc <- 1
    alpha[i] <- alpha[i-1] + runif(1, min=-.5, max=.5)
    if (alpha[i]<=0) {acc=0; alpha[i] <- alpha[i-1]}
    if ((lnpost(alpha[i], theta[i], x) - LL) < log(runif(1))) {
      alpha[i] <- alpha[i-1]
      acc <- 0}
    acca <- acca+acc
    if (i%%100==0) {print(c(i, alpha[i], theta[i], acca/i, accq/i))}
    }
amean <- mean(alpha[n.burnin+1:n.iter])
qmean<-mean(theta[n.burnin+1:n.iter])
Fit <- structure(list(alpha=alpha, theta=theta, amean=amean, qmean=qmean))
return(Fit)
}

The above code will just hand off the parameter sampling between alpha and theta (which is the scale and not rate) and will, hopefully, eventually settle down. I added two extra parameters for iterations and burnin, and eventually, I could add another, outer, for-next loop for multiple chains.

The next code takes advantage of Gibbs sampling for the scale parameter:

MHGamGib <- function (x, n.iter=10000, n.burnin=0){
alpha <- matrix(.5, n.iter)
theta <- matrix(10, n.iter)
acca<- 0
lnpost <- function(a, q, z){
  NLL <- (a-1)*log(z)-(z/q)-a*log(q)-lgamma(a)
  return(sum(NLL))
}
for (i in 2:n.iter)
{
#sample theta
  theta[i] <- 1/rgamma(1, rate=sum(Test), shape=alpha[i-1]*length(x)+1)
#sample alpha
    acc <- 1
    alpha[i] <- alpha[i-1] + runif(1, min=-.5, max=.5)
    if (alpha[i]<=0) {acc=0; alpha[i] <- alpha[i-1]}
    if ((lnpost(alpha[i], theta[i], x) - lnpost(alpha[i-1], theta[i], x)) < log(runif(1))) {
      alpha[i] <- alpha[i-1]
      acc <- 0}
    acca <- acca+acc
    if (i%%100==0) {print(c(i, alpha[i], theta[i], acca/i))}
    }
amean <- mean(alpha[n.burnin+1:n.iter])
qmean<-mean(theta[n.burnin+1:n.iter])
Fit <- structure(list(alpha=alpha, theta=theta, amean=amean, qmean=qmean))
return(Fit)
}

Note that I'm still parameterizing using the scale by sampling the rate and using its reciprocal.

To set up, run, and compare the models, I used the following code:

set.seed(22)
 
Test <- rgamma(100, shape=2.5, scale=1e5)
 
MHT<-system.time(MH<-MHGam(Test, n.iter=50000))
 
MHGT<-system.time(MHG<-MHGamGib(Test, n.iter=50000))

The Gibbs version runs about 40% faster, comparing MHT with MHGT (at least on my laptop), which stands to reason as Gibbs sampling is Metropolis sampling where every variate is accepted. What is more fascinating is how fast the parameter values converge:

Using the following code to overlay the trace plots for theta:

plot(MHG$theta, type="l", col="red")
points(MH$theta, type="l", col="blue")

We see:

The Gibbs sampling (in red) gets to the right neighborhood much more quickly for theta than the Metropolis (blue), and gives a more robust (less serially correlated) answer.

What is also surprising is how the better estimates for theta affect the random-walk Metropolis algorithm for alpha. Using the following code:

plot(MH$alpha, type="l", col="blue")
points(MHG$alpha, type="l", col="red")

We get the following trace:

Personally, I found that, writing the algorithms and seeing the direct results helped me understand the concepts better than just reading, and I'll certainly try to use Gibbs sampling where I can!

Social media redux

2011-06-07T13:24:00.000-04:00

I'm certain that by now, everyone knows about Congressman Anthony Weiner, his social media behavior, and yesterday's press conference. I'm not going to discuss Weiner's immoral and, dare I say it, despicable behavior in his online actions and the ensuing attempt at a coverup. What I do want to mention is that this serves to reiterate one of the important points I mentioned in yesterday's blog post: that anything you say or post is there forever. If you would not want something on the front page of your local tabloid, don't post it anywhere on the Internet. Of course, if you always attempt to behave appropriately, chances are you won't get into such a mess to begin with, but as we say in the business - that is out of scope for this discussion.

Social Media guidelines specific for (re)insurers

2011-06-06T14:31:00.001-04:00

With social media here to stay, and many of us (actuaries, cat modelers, and other (re)insurance company personnel) actively involved in the social media on both personal and professional levels, it behooves us to remember certain social niceties. Everything that you post on the Internet is pretty much there for posterity; 50 years from now, the wayback machine will very likely be able to retrieve that one time you posted a picture of yourself at the holiday party—after spiked eggnog #7. My personal, primary rules, whether it be on twitter, linkedin, facebook, wikipedia, or some random bulletin board are:

Never post anything on-line that you would be uncomfortable saying to someone's face in public.
Never post anything on-line that you would be devastated seeing on the front page of the New York Times.

I am all for having a certain amount of silliness and fun on-line, I mean, where are you going to let your hair down, at work? However, there is a big difference between good-natured silliness and flaming. Too many people believe that there is anonymity on the Internet, and thus act selfishly, spitefully, and all-too-often just plain cruelly with the mistaken assumption that they will never be uncovered. Besides for this being awful and immature behavior, which speaks loads about the person demonstrating it, it can be a serious operational risk hazard for the person's employer, and thus potentially grounds for termination for the employee. For example, it has not even been a year yet since a Massachusetts schools' supervisor lost her job over a facebook post. To demonstrate Santayana's maxim still holds, a teacher in Brooklyn did not learn from history, and seems to be in danger of losing her job for the same reason.

Mairi Mallon, founder and managing director of rein4ce, and @reinsurancegirl on twitter, has recently authored a detailed post on her blog about this very topic (from which I shamelessly stole the idea). I very highly recommend anyone even tangentially involved in social media read it. Better yet, subscribe to her blog (which can be found in the blogroll at the side of this post). I think her summary is important enough that it bears repetition:

Know and follow our (INSERT LINK TO OWN CORPORATE GUIDELINES) corporate guidelines . The same rules apply online.

Users are personally responsible for the content they publish on blogs, Facebook LinkedIn, Twitter or any other form of user-generated media.

Identify yourself—name and, when relevant, role within the organisation—when you discuss company or company-related matters. You must make it clear that you are speaking for yourself and not on behalf of the company.

Respect copyright, fair use and financial disclosure laws.

Don’t provide our or another’s confidential or other proprietary information. Ask permission to publish or report on conversations that may be deemed to be private or internal to the company.

Don’t cite or reference clients, partners or suppliers without their approval. When you do make a reference, where possible link back to the source.

Don’t use ethnic slurs, personal insults, obscenity, or engage in any conduct that would not be acceptable in our workplace (Mallon 2011).

Engaging the social media can be exciting and rewarding, both on a personal and a professional level; make certain that you don't hurt yourself, your image, your integrity, and your future when you do so. Everyone wants to be the next Richard Branson; I'm less certain about who wants to be the next Aleksey Vayner.

References:

Mallon, Mairi, “Social Media guidelines — a freebie for insurance and reinsurance bods.” Weblog entry. Reinsurance girl's blog. June 6, 2011. June 6, 2011 (http://www.rein4ce.co.uk/blog/?p=315).

Solvency 2, what I don't understand

2011-06-05T15:49:00.000-04:00

There has been much fanfare about Solvency II, and how it will be coming to the US now that it is pretty much fait accompli to be a requirement for UK and European insurance companies starting in 2013. Perhaps I am missing something basic, but blow are some of the issues that I have with Solvency II:

Isn't Solvency II based on the same principles as Basel II/III? Now how well has that worked for Europe. Greece, Iceland, Ireland anyone?
The first pillar of Solvency II (Article 75 (1)(b)) requires that "...liabilities shall be valued at the amount for which they could be transferred, or settled, between knowledgeable willing parties in an arm’s length transaction." I don't know about the rest of you, but there really is no liquid market for (re)insurer loss reserves. I've been involved in pricing a few loss portfolio attempts--none of which came to fruition mind you--and each one is really a bespoke transaction. Different counter-parties to the same transaction will arrive at different values for the reserves. So how is Solvency II going to handle this?
At its heart, the capital requirement is still a Value at Risk (VaR) measure, albeit at the 99.5%-ile. When will people learn that VaR is a rather non-robust statistic? It is a point on the cumulative frequency distribution, with no "knowledge" of what is above or below it. It is very susceptible to discontinuities (think step function), and its components are non-additive. At the very least, the measure should be based on TVaR, or the expected value above a given point. As an expectation, it is additive in its components (Co-TVaR measures exist and are meaningful) and it, as a first moment, reflects to some extent the entire distribution in the tail above the point, not just a point. As we all know, actually seeing any particular result from a continuous distribution almost never happens, which is why we talk intervals and not points.
While Operational Risk is certainly a significant factor in a company's risk profile, I have yet to see any good measure or process to quantify the expected value of said risk.

Unlike the banking industry, the insurance industry, especially in the US, has been rather stable for the past century or so, and while I agree that certain elements of Statutory Accounting, such as the holding of nominal reserves, may seem somewhat extreme with respect to other businesses, it has done rather well in ensuring a viable and solvent insurance industry. Why are we rushing into something which has neither the provenance (Basel) nor the track record (not even officially implemented in Europe) to justify adding a new accounting and solvency standard.

R-bloggers

2011-05-18T20:48:00.000-04:00

As I decided to try and blog a little more often now, and touch on "R" every now and then, I decided to take R-bloggers up on their standing offer to include R-related feeds at their site. So, everything I tag with "rstats" (you can guess where that came from) should flow through to them. I've added them to my (tiny) blogroll at the side of the blog, but if you just cannot wait to see what they are all about, and I recommend it, you can just go here.

Why method of moments doesn't always work

2011-05-15T20:37:00.001-04:00

A number of years ago, someone asked me "why does my company need actuaries to fit curves, once I have the mean and standard deviation of my losses, isn't that enough?" I explained to him that not every distribution is completely determined by its mean and standard deviation (as the normal and lognormal are), and as at that point, I did not have "R" installed on my laptop, I demonstrated it to him in Excel. Having wanted to start blogging about "R", even ever so infrequently, I figured I'd toss together a little code to demonstrate.

The example I gave was to compare a gamma and a pareto distribution, each of which has mean 10,000 and a CV of 150% (making the standard deviation 15,000). I will spare all of you the algebra, but suffice to say, that using the Klugman-Panjer-Wilmot parameterization (which is used by most casualty actuaries in the past 20 years or so) the parameters of the gamma would be theta (R's scale) = 22500 and alpha (R's shape) = 4/9. The equivalent pareto would have theta (R's scale) = 26000 and alpha (R's shape) = 3.6.

Graphing the two (and Hadley, please forgive me for using default R' plotting, I left my ggplot book in the office; mea culpa) you can easily see how the distributions are rather different.

To make things easier for me, I used the actuar package to do the graphing:

library(actuar)
curve(dpareto(x, shape=3.6, scale=26000), from=0, to=100000, col="blue")
curve(dgamma(x, shape=4/9, scale=22500), from=0, to=100000, add=TRUE, col="green")

Created by Pretty R at inside-R.org

Obviously, the tails of the distributions, and thus the survival function at a given loss size, is different for the two, notwithstanding their sharing identical first two moments. So, this was just a brief but effective visualization as to how the first two moments do not contain all the information needed to find a "best fit," and why we like to use distributional fitting methods (maximum likelihood, maximum spacing, various minimum distance metrics like Cramer-von Mises, etc.) to get a better understanding of the potential underlying loss processes.

New look!

2011-05-13T14:42:00.001-04:00

I decided to change things around a bit, and applied a new template and color scheme. Maybe now my next post won't take another five months! Let me know if you like the new look.

It has been a while

2011-01-12T10:39:00.002-05:00

It has been quite a while since I posted an update here. Possibly, because the longer I work, the less time I seem to have. Just to share something, I did recently have some fun programming a random gamma variate generator in VBA for some copula work. While I'm certain there are much more elegant implementations out there, it was good to figuratively roll up my sleeves and dust off some VBA programming skills. Now I can keep that line on the resume!

IE6 no more

2009-09-22T17:07:00.003-04:00

I just came across this article, entitled "IE6 No More! Popular Web Companies Start Project to Kill IE6" and decided to add the code to my blog, so if you are using IE6, you may see a note to that effect. FireFox is the way to go anyway!

Hey, that sounds like my studying routine!

2009-09-21T16:49:00.003-04:00

In my daily perusal of the intellectual side of the news, the comics, I came across another gem, which while ostensibly applied to elementary school, may be the best actuarial study technique ever printed. Enjoy!

http://d.yimg.com/a/p/umedia/20090920/largeimage.7b0f57af9a54b45f31e113377789cccb.gif

Actuaries in the media…

2009-09-10T16:49:00.004-04:00

…well at least in cartoons. Unfortunately, even we are not immune from politics.

http://d.yimg.com/a/p/uc/20090910/largeimagepo090909.gif

Future Education Methods and the Actuarial Profession

2009-09-09T12:09:00.004-04:00

Recently, the "big three" actuarial organizations in North America, the CAS, SOA, and CIA, have promoted a proposal to radically change the actuarial education methods used to credential actuaries in the United States and Canada. Their letter describing the suggestions may be found here.

Personally, I think it is a very bad idea, and I sent the following to the feedback address at fem@actuarialdirectory.org. The deadline for responses is tomorrow, so if you have not already commented, you have one more day.

I believe that the proposed future education method (FEM) of exempting actuaries from examinations based on passing specific university courses is a bad idea, and I am strongly opposed to it.

Currently, one of the strongest elements of our profession is its being close to a true meritocracy; much more so than any given university. A candidate takes an examination simultaneously with his or her peers, has his or her exam anonymized, and then graded by anonymous referees (anonymous to the candidates) similar to a double-blind test, and has his or her results, still anonymous, compared with the results of all other test-takers. Then an pass score is set (we will leave the discussion of the score being set based on the actual distribution of results to another time) and all who have met or exceeded the score will be deemed to have net the requirements. There is very little opportunity for either "gaming the system" or for allowing any other external influence (pernicious, malicious, or otherwise) to effect the outcomes. Contrast the above with a university setting. Notwithstanding the various societies' intentions of ensuring "specific standard", it is undeniable that different teachers, and different universities for that matter, will have different standards. Furthermore, this opens the hideous potential for student-teacher relationships impacting the results, be they for good or ill. We lose the anonymity, we lose the simultaneity, we lose the ability to compare oneself to one's peers, we lose the "double-blind" nature of the exam process.

Another distinct shortcoming of the new process is the complete loss of standardization. Currently, all who pass Exam "X" in a given sitting do so with the exact same exam. Symmetrically, all those who fail Exam "X" in a given sitting, fail the same exam that those who pass have passed. When a change is made to the syllabus, and thus the exams, it will affect all sitting candidates in the same way. There is no preferential treatment possible. Under the new system, there can be many different "exams;" as many as there are approved courses. The syllabus, and any changes thereto, will not affect all candidates equally, but is completely dependant on the institution, and for that matter the teacher. This also includes the potential dilution of candidates, in that their passing is no longer controlled by the respective societies, but by external institutions.

A third distinct shortcoming of this new process, while less egregious, is the loss of its screening abilities. As a former employer of mine, a credentialed actuary, once told me in half-jest, "you do not pass actuarial exams, you win them." One of the benefits of the exam process, and I say this in hindsight having completed it, is its ability to isolate people who can deliver good answers under pressure. I cannot speak for anyone else, but to me, this is pretty much the epitome of what we as actuaries are called upon to do. We are not supposed to always have the best answer, but we are supposed to be able to help our clients (be they actual clients, employers, etc.) to address their concerns as best we can in the ever-shortening time-frames we are given. Furthermore, in actual work, we are not always given all the information "handed" to us on a platter. We need to be able to quickly identify good and useful data from bad, and use it as best we can. Exam work is very similar to this. We have to learn the material on our own—exam seminars are good refreshers, but I have never met an actuary who has only used them without other studying who has had any success—which includes triaging the material, identifying the areas on which to focus (the Brosious quartering comes to mind), and quickly "ingesting" what is needed to achieve success. In a university setting, much of this is lost. For all we know, the final can be a one week, open book, take home affair—unthinkable under the current method.

Regarding the purported benefits of such a system, I must respectfully disagree with some.

I do not see the quality of actuarial education improving any more because of institutionalized exam exceptions than it would should the current syllabus and exams be appropriately updated. Some of our syllabus papers, at least when I was a candidate, were written not for candidates but for actuaries, and experienced ones at those. What we need is a centralized effort at generating better exam material, not diffusing that effort throughout 20 or so universities, each with their own agenda and intentions.

The following, I feel, highlights a critical chasm between academic actuaries and practicing actuaries. I would like to know why "university" training in actuarial methods is considered a benefit? Many would argue that in truth it is a detriment. Universities are often known for their being "out-of-touch" with what is occurring in the "trenches" as it were. The ivory towers of academia are bastions of theory, but oft-times are highly impractical. As actuaries, most of us will practice, and not remain theoreticians, and as such, I would counter that one of the most important elements in becoming and being an actuary is the "apprenticeship" one undergoes in one's early actuarial career. Knowledge should never be confused with, or substituted for, experience, and it is experience that is key in our discipline, which at times is as much an art as it is a science (at least in my field of casualty reinsurance).

I can understand how allowing university classes to exempt students from actuarial exams would increase the career's attractiveness. However, that can be addressed by having the university courses geared towards the specific exams and preparing their students for taking said exam.

In closing, I feel that the FEM described would do far more harm than good to our profession, and I urge the societies to reject it in favor of enhancing our current process.

Thank you.

Cassini photos of Saturn and its environs

2009-04-21T00:10:00.002-04:00

Cassini photos of Saturn, its rings, and its moons, as seen here http://www.boston.com/bigpicture/2009/04/cassinis_continued_mission.html, are simultaneously amazing and awe-inspiring. It is further serendipity that I am currently reading Jack McDevitt's Academy series; the one where the first book features the monument on Iapetus, whose surface is captured in glorious detail by Cassini.

Another bit.ly test

2009-04-06T10:23:00.001-04:00

Hmm, the API key didn't work. One more try.

Bit.Ly test

2009-04-06T02:18:00.002-04:00

Now that I have twitterfeed posting to ping.fm, and from there to twitter and identi.ca, I'm trying one last time to get a shortened URL that does not contain "ping.fm" in it, for people behind websense or the like. This post will check the activity of a bit.ly account (not anonymous) as the shortener in twitterfeed. Let's hope this works!

Twitterfeed Test

2009-04-05T20:37:00.001-04:00

Well, twitterfeed through Ping seemed to update both Twitter and Identica, but the url shortener has a ping domain which is blocked at work. Trying something else now. This may be a many-post day trying to get this set up.

Another test

2009-04-05T12:11:00.003-04:00

Well, it seems that Ping.fm is great for cross-posting my tweets to identi.ca but it didn't take the previous blog post. So, let's try Twitterfeed feeding Ping.

Ping.Fm and the social-interaction revolution

2009-04-05T09:56:00.001-04:00

We are witnessing a social-interaction revolution. First blogging, and then micro-blogging, have taken cyberspace by storm. With so many different options and niches, it may be difficult to manage all of one's separate service providers. There is a service, Ping.Fm, which promises to handle that. You can set it to accept feeds from one service and cross-post them to others. You can specify three different kinds of feeds (status, blog, and microblog) to decide which goes where; you can set up special tags to post to s defined subset of feeds, or you can even specify a particular feed. This blog entry is being made to test if Ping.fm is going to handle blogger posts the way I expect. Enjoy!

A single currency?

2009-03-31T00:39:00.005-04:00

On Bloomberg Surveillance this morning the concept of a single world-wide currency was discussed. Personally, I am not sure how that would even be possible. How would investors or the market be able to reflect their opinions as to the relative security and stability of various governments?

Even with a single world-wide currency, governments and countries will have different levels of output, stability, production, etc. There needs to be a way for the markets to reflect this. Currently, this can be done by the relative value of one currency versus another. Now consider every country has just one currency. The governments, natural resources, and indigenous economies are still different. Unless the currency is 100% backed by some universal commodity (gold, for example) part of the value of the currency is the "full faith and credit" in the issuing body, and that the currency note is redeemable for equivalent goods and services across national borders. That discrepancy between entities still exists regardless of shape, color, size, and portraits on the bills or coins in use.

Perhaps the discrepancies will become evident in different interest rates that need to be offered by the treasuries of the respective countries; a government with a stronger economy can offer lower rates in return for the inherent stability. Well, doesn't that make government securities in and of themselves a "shadow currency"? We can have the two bonds with the same par value, the same coupon, and the same duration with different prices, due to the perceived difference in risk and stability. I'd guess that pretty soon thereafter a market for trading these identical yet different instruments will be created, and we are pretty much back where we started.

As this is just semi-knowledgeable ramblings on my part, I'm curious to know how investors bet the German economy against the French, if both use the Euro. While an optimal currency area may be larger than a country, I'm pretty sure it is smaller than the entire world.

AIG or ARRGGHH?

2008-06-18T00:46:00.002-04:00

Interesting news today, from CNBC:

American International Group CEO Martin Sullivan stepped down from his post, as expected, after growing discontent among the board and shareholders with his management of the insurance giant.…Sullivan has come under intense criticism since AIG recently announced billions of dollars in writedowns from losses after Sullivan and the company assured investors the writedowns would be minimal.

This is another example of where a good ERM study would have exposed the asset weakness. As The Economist writes:

Mr Sullivan did himself no favours by mismanaging expectations. He and his team poo-pooed critics, insisting until this year that AIG had oodles of excess capital and that its actual (as opposed to mark-to-market) losses would be modest. The firm has since been forced to admit that its accounting models were too optimistic, after its auditors found “material weaknesses” in them

Perhaps an outside analysis of the capital model may have been a good idea?

MA INSURANCE COMMISSIONER PROPOSES BANNING USE OF SOCIOECONOMIC FACTORS IN RATING AND UNDERWRITING

2007-09-04T13:37:00.000-04:00

From Insurance Journal - August 29, 2007

The draft regulation makes years of driving experience, driving record and a vehicle's model and safety features as the primary rating and underwriting factors in private passenger auto insurance.

The draft regulation bans insurers from using gender, marital status, education, occupation, national origin, religion, homeownership and other socioeconomic factors, some of which Burnes says are prohibited by statute and others which she deemed violate public policy, for either rating or underwriting.

It also forbids insurers from using information from credit reports or scores for rating for at least a year, during which time she said she would study the overall use of credit scores by insurers. It does not prohibit the use of credt scoring for underwriting.

This does mean that subsidization will still be alive and well. Let's hope that there is less of it.

Florida turbulence

2007-08-20T12:47:00.000-04:00

At least some Florida legislators are seeing the light. Florida Underwriter writes:

The resulting 91-page bill swept through the insurance code with all the fury of a major storm that left insurance representatives stunned and some legislator’s in disbelief. Representative Dennis Ross (R-Lakeland) said the bill was “as close to a socialist policy that we ever came to.” He was joined by Representative Don Brown (R-De Funiak Springs) who decried the changes to Citizens Property Insurance Corporation, which makes it certain the former residual market will be the state’s largest insurer. Speaking of the freeze of Citizens’ rates until January 1, 2009, Brown noted that the bill ensured that homeowners would be eventually face potentially crippling assessments. “A one-year rate freeze is playing Russian roulette and every year you continue the freeze you’re adding another bullet to the chamber,” he said.

Leave it to Governor Crist, however, to never confuse reality for political expediency:

But no matter how dire the message, Crist and the legislature were more than happy to press forward and place their imprint on the property market. “The Florida House and the Senate have been working very hard to bring relief to our homeowners, and I am grateful for their dedication,” said Crist . “The people have pleaded for relief, and the good men and women of the Florida Legislature have answered the call. This is the right thing to do.”

The right thing to do? For whom?!

Workers Compensation Data

2005-08-01T11:45:00.000-04:00

This is interesting, and useful. Needing to do research on workers comp rate filings, I stumbled on this site, which links to near every rating bureaus you can think of: http://www.cutcomp.com/depts.htm

Rather helpful.

The Further adventures of a failed behemoth

2005-08-01T11:08:00.000-04:00

From the Insurance Journal: August 1, 2005

Reinsurers Sue AIG Alleging Claims Fraud; AIG
Denies All Wrongdoing
A group of 18 insurers is suing American International Group and bankrupt
fronting company Trenwick America Reinsurance Corp. for allegedly scheming to
collect as much as $73 million in what the insurers claim are "grossly inflated"
workers compensation and other reinsurance claims.

The suit alleges that only about $15 million of the $73 million in claims for which AIG has demanded payment appear to be legitimate paid losses eligible for reinsurance coverage. The remaining amount reflects "highly suspect" estimates of future (incurred but not reported) losses, according to the complaint, which was filed in Suffolk Superior Court in Boston on July 6.

Click on the above link for the rest of the article.
It is amazing to see what is happening with AIG. It is even more amazing to see just how much AIG & Co. was doing sub rosa. I am very curious to see what else is unearthed. Stay tuned, this is going to develop for a long time.