Quite a bit has been written recently on securities litigation trends but very little of it is especially enlightening, says Jim Franson, who discusses how re/insurers can evaluate and use the information

Within the abundance of written commentary, there appear to be materially different theories in predicting the future development of securities claims - even when those commentators are basing their conclusions on the same source material. The three most commonly-cited sources of information on these trends - Cornerstone Research, NERA and PwC - are all well-researched and academically sound reports, but a reader legitimately could wonder if the loss development trends are improving or deteriorating, and seemingly could find support for either conclusion. For example, reviewing information on the 2003 year shows that the number of actions filed in 2003 declined slightly over 2002 but the amounts of settlements increased.

This confusion is not the fault of the analysis. Rather, it reflects a fundamental problem with the information itself: the statistics simply do not lend themselves to unhedged conclusions or to tidy examination.

So how do we make some sense of the problems associated with this analysis, and how can we evaluate and use this information for our own predictive needs?

In order to start and make the information relevant to re/insuring these risks, it's important first to understand some fundamental problems with our information and processes. Unlike property insurance and shorter tail casualty lines, professional liability, and directors' and officers' liability (D&O) claims in particular, are relatively long-tail in nature. We will generally not know for a period of years the effects of legislative changes, litigation trends, economic shifts (for example, stock market indices) and, of course, changes in insurance pricing.

Put simply, it is difficult to tell whether a given underwriting year is profitable and then, not for quite some time. Moreover, the constantly shifting sands of the market forces at work result in very uncertain footing for any predictive analysis. And finally, our awareness of the failures of several insurers and reinsurers with significant involvement in these areas is a constant and painful reminder of the small margin for error within which capital providers expect re/insurers to operate.

Parameter risk

In treating D&O and professional liability risk as a relatively long-tail line of business, underwriters make a variety of assumptions before assuming that risk. For example, we make assumptions about frequency, severity and pricing, and overlay these variables on top of other assumptions about profitability. What makes this task more difficult is the array of external forces affecting these variables.These variables may be market-based (such as competition), regulatory, legal or legislative, or other largely-unpredictable forces. And these variables change constantly.

It often takes years to determine whether a portfolio is profitable, and so there are numerous opportunities for a single incorrect assumption (or series of assumptions) to turn a profitable position into an unprofitable one. In a low interest rate environment where investment income is very modest, there is very little room for error.

Two other problems are also relevant. I call them the "survivorship bias" and the "as if" problems.

Survivorship bias

One of the fundamental systemic problems with our data is the existence of a survivorship bias. Here's the problem: the historical underwriting data that we analyse is usually limited to healthy insurers still in the business (this is less of an issue when evaluating a single transaction, but is a real concern when evaluating an overall line of business.) Companies and portfolios that have 'blown-up' are usually removed from the population that we evaluate - and so we are conditioned to evaluate only what is right before us, ignoring what is not readily evident. But in order to fully consider the relative profitability of a line of business, don't we need to see the whole picture?

Unfortunately, this data is very difficult to obtain and to confirm its accuracy when we do get it. But wouldn't it be helpful to view results by adding back in the portfolios of companies like Reliance, Kemper or any of the other companies that have been acquired or absorbed by larger, healthier companies? Presenting a more complete picture permits us to establish a more accurate baseline. Only from there can we evaluate changes in frequency, severity and pricing.

As if

The as if problem is similar to the survivorship bias, but considerably more prevalent. When presented with portfolios that have had some tough experience, it is common practice to identify the source of the problem and then scrub the data to remove the source. The result is a more attractive portfolio. If this represents a class of business that will no longer be considered (ie "we no longer underwrite investment banks"), this as-if approach is valid, but the original experience should still be considered.

Perhaps the offending line of business is indeed permanently carved out, but it is still helpful to view how the results were affected by a single class or trend, providing valuable insights should a similar, but unrelated trend surface in the future.

However, the as-if phenomenum is not limited to portfolios of insurance.

Reports on securities class action trends are now being 'adjusted' to 'back out' unusual events such as laddering or mutual funds. The intention is to capture trends on traditional cases, but one of the key takeaways lost in the data manipulation process is that there is nothing 'traditional' about this class of business. The unusual events still represent market losses and claims that some segment of the insurance industry will have to pay.

Another area where inconsistent data clouds predictive analysis relates to relative profitability calculated by evaluating variables such as claims trends. Results are interchangeably presented as 'calendar year', 'accident year' and 'underwriting year'.

We need to pay special attention to these differences because significant environmental changes are simultaneously at play, and an innocent error related to which 'year' is at issue can produce dramatically different observations. For example, the 2002 accident year (or report year for some D&O insurers) arguably was a very bad year for a number of D&O insurers and reinsurers. Numerous very significant claims erupted. However, these claims may have attached to policies written in 1999, 2000 or 2001 (remember this is claims-made business), and at that time, many multi-year policies were still in effect.

In contrast, the 2002 underwriting year should have markedly different characteristics. Rates had been increasing, claim activity appears to be down slightly compared to the previous year, and policies had predominantly one-year terms. While the 2002 underwriting year remains immature, one may reasonably conclude that it should perform dramatically differently from the 2002 accident year. Additionally, other changes must be considered: coverage grants (for example, post-1995 entity coverage), changes in business mix (is the portfolio more weighted to one sector and changing over time?), as well as changes in the limit and attachment profile (affecting severity).

The as if problem is the most difficult issue to contend with. Past systemic problems such as the S&L crisis in the 1980s, IPO laddering, research claims, Enron/Worldcom corporate scandals, and most recently claims involving mutual funds, are unlikely to repeat themselves in the same form. Nevertheless, extreme events do occur, they have occurred recently, and they will happen again, albeit under a different name.

Yet the re/insurance industry is preconditioned to strip out these anomalies, reasoning that these are not everyday occurrences and will not happen again. Rather, we focus on traditional types of losses and when we evaluate data, such as securities class action trends, we seek to normalise the data to make it fit with our preconceived notions of what the losses should look like. When analysing losses, we are confronted with a 'stationarity' problem - the (incorrect) assumption that the properties that affect loss costs are either stationary or constant. They are not. Insurers and reinsurers need to pay attention to data on traditional losses when making decisions about frequency and severity, but we must also recognise that the very nature of this business has catastrophe characteristics and, as such, pricing should contemplate these extreme events.

The data

When examining the most recent findings, it is important to bear in mind that any predictive analysis necessarily represents a 'point-in time' evaluation. Claims trends, legislative trends, legal activism and economic forces simply do not follow a tidy calendar year. There is no schedule for when changes will occur or how long they will last.

We are also in the midst of a dramatic change in the overall loss landscape.

Not only are we still trying to evaluate the effects of the Private Securities Litigation Reform Act (PSLRA), but now we will have to determine the likely impact of the recent Sarbanes-Oxley legislation. Will this new legislation add volatility? Probably. Will it make it easier for the plaintiffs' bar to bring cases? Quite possibly.

One clear conclusion that we currently can make is that post-PSLRA cases are taking longer to settle. In the pre-PSLRA era, 61% of cases were resolved (either by settlement or judgment or dismissal) within three years, with 77% resolved within five years. In the post-PSLRA environment, only 44% of cases were resolved within three years, and only 62% were resolved within five years. Because of the amount of cases that remain unresolved - and without knowing what amounts those cases ultimately will settle for - there exists the potential that the economics of the business are become much more speculative. At a minimum there is that much more uncertainty created by the lengthening of the tail.

The current statistics tell a confusing story, and without sufficient understanding, there is the danger of making the wrong conclusion. For example, settlements expressed as a percentage of investor losses have dropped from 7.2% in 1996 to 2.7% in 2002, increasing only slightly to 2.9% in 2003 but still down overall. One might conclude that severity is abating, but that would be incorrect. Not only are market capitalisations higher, but these are largely 'resource based' recoveries, particularly in the larger cases. When a company is sued for over $1bn, but carries only $100m in insurance, chances are very high that the case will get settled for about $100m. This creates the illusion of less severity, but the insurance dollars involved have not decreased. Looking at figure 1, which shows securities class action filings by year, 2003 shows a reduction in filings over 2002. That appears to be a favourable trend. And even when viewed against the pre-1996 period, the 2003 figure looks to be in line.

But a surprising contrast is that, based on these filings (and adjusting for the number of publicly-traded companies), the probability of a company being sued rose approximately 40% from 1995 to 2002 (source: NERA report 2003). Both observations are factually correct but can be applied in markedly different ways when used as a decision-making aid.

When looking at the most recent settlements, comparing 2003 to 2002, the median settlement value fell by 12.5%. However, the average settlement value rose 19.6% for the same period. The trends thus appear to be moving in different directions. How can that happen? The reality is that the 'traditional' cases appear stable, but the average is skewed by large settlements. Stated differently, this part of the data suggests stable frequency, but increased severity.

But perhaps even this is too simple an answer. Legal experts in this arena suggest that the plaintiffs' bar uses the smaller, traditional cases to generate regular cash flow, with a view toward funding the larger, more time-consuming and expensive cases they must prosecute vigorously.

These are the cases that make for headline news and, in turn, drive the average settlement values higher. Viewed another way, the claims environment is bifurcated: one section comprised of traditional cases, and the other being made up of the high-severity 'blockbuster' cases. This school of thought would conclude that the unusual blips or anomalies will remain a part of the landscape; and that those cases we had considered 'outliers' may now become more usual.

When analysing this data, it is critically important to view not only the measures of location (median and mean), but the measures of dispersion (skewness, standard deviation, kurtosis) as well. This gives us a more complete picture of the volatility and these measures in this case are relatively high. As an illustration, if we take the 2004 new case filings on a year to date basis and then project them out for the full calendar year, we obtain a projection of approximately 228 new cases - which would suggest a stable new claim environment. But if we use a simple least squares method to trend forward, we obtain a projection of 300 cases - a 30% difference from our earlier projection and a 40% increase over the 2003 figure. Both of these projections suggest an increase over 2003. Neither of these scenarios may play out, and they are far enough apart that we should not be overconfident and assume that we know where things are headed.

There are other reasons to believe that the extreme or outlier cases will continue to be filed and prosecuted to settlement. A new breed of activist attorneys general and the actions that they bring (so far, research analyst, mutual funds and possibly insurance brokerages) has asserted its presence. The trend toward more aggressive Securities and Exchange Commission and Department of Justice investigations and settlements is another area. Although these actions appear to have peaked in 2003, the conventional wisdom suggests that these bodies have simply got more efficient about the cases they bring. Sarbanes-Oxley provides for a more active climate of enforcement, and the recent large settlements create a war-chest for the attorney general legal staffs in various states to pursue new cases.

Looking around the corner

Re/insurers need to recognise that we are operating in a different environment than in the past. The rare event fallacy and our tendency to remove those losses that we consider outliers need to be reconsidered as many of these types of losses have the high potential to be repeated. The information in the periodic analyses of securities litigation trends is definitely valuable. But it is very easy to over-emphasise any one particular data point and make a wrong inference. It is more important to take a step back and carefully evaluate the entire situation, and even then do so with the full knowledge that this is unpredictable business and anything can happen.