Big data and actuaries – are we part of the solution?

Investigating Big Data

Investigating Big Data

The Actuaries Institute had a financial services forum last week, which I managed to get to some of. My first report is of the Big data plenary session which closed the conference.

I’ll start with some background reading:

Cheerleaders for big data have made four exciting claims, each one reflected in the success of Google Flu Trends: that data analysis produces uncannily accurate results; that every single data point can be captured, making old statistical sampling techniques obsolete; that it is passé to fret about what causes what, because statistical correlation tells us what we need to know; and that scientific or statistical models aren’t needed because, to quote “The End of Theory”, a provocative essay published in Wired in 2008, “with enough data, the numbers speak for themselves”.

This is Tim Harford  recently in the Financial Times, questioning much of the breathless prose that surrounds big data. Harford goes on;

Unfortunately, these four articles of faith are at best optimistic oversimplifications.

Read the rest of the article, but Harford points out that having lots of data is no substitute for trying to understand its strengths and weaknesses (particularly in selection bias) before you analyse it. That’s a lesson that most actuaries have had to learn at some stage, and hopefully will not unlearn too quickly

At the FS Forum, there was a bit of breathless excitement, but some sensible commentary too. In a session with many quotable moments, one particularly stuck out for me, from Duncan West:

Knowledge is power. Those who understand the performance and value drivers of their business outperform those who don’t.

Fundamentally, Harford and West’s points are similar. Understanding your business and its drivers of value has always been the way to outperform. Throwing data at the wall and hoping it will stick (which is the kind of thing outlined in Harford’s articles of faith) won’t necessarily improve your understanding of those value drivers. But having a superior understanding of your business through transforming the right data into information all the way along the value chain can make a big difference to your performance, not just for shareholders, but for customers too.

Several speakers throughout the conference asked the question of whether life insurers have under-invested in data compared with general insurers in Australia. From my experience of both, I would say yes, that is the case. But if you compare either life or general insurers in Australia with the world’s best practice companies at using data to transform their business, there wouldn’t be many insurers who do well out of the comparison, which is perhaps a bit of an indictment of our profession, if we really are playing the role of data scientists in insurance companies.

One question from the floor was what actuaries should do about it. After all, there have been a few articles that explicitly link actuaries and data scientists – for example this one from the SOA that explicitly claims that “actuaries are data scientists for insurance”. Is this true? Have actuaries played that role? And if so, how well have we done it?

The best actuaries in insurance companies can interrogate data in ways that inform product managers, claims managers and underwriters how to select, price for, and manage their risks and business. And the insurance business is not only about insurance risks, but business issues like retention and customer satisfaction. The control cycle framework, if you learn its lessons, is a great way to think about the whole end to end profit cycle.

But it is fairly easy for an actuary who enjoys interrogating data to get lost in analysis for its own sake. To torture the data until it confesses. To spend all their time getting the right answer, without it necessarily being the useful answer that will help the business make better decisions. The stereotype of the actuary who sits in the corner with spreadsheets did not come from nowhere.

So how should actuaries be responding to big data?

The answer depends on what kind of actuary you are. I’m paraphrasing Duncan West here, but his practical suggestions were good ones.

  • If you are in a leadership role in your company, get data onto the strategic agenda. Many companies talk about the importance of data, but talk is cheap. Do they manage themselves in ways that demonstrate the importance of understanding value drivers? If not, what should they be doing strategically?
  • Actuaries in any role need to go away from a regulatory and compliance mindset that accuracy is the most important way to measure success. They should measure success by helping the business to make good decisions.
  • And actuaries at all levels need to develop the skills necessarily to show insights to the business. Insights are useless if the business can’t understand them. So communicating insight is a key part of an actuarial role.

And don’t try to change the world with a massive IT solution without thinking about how the business works. Incremental steps are more likely to succeed and gradually change the culture to get data onto that strategic agenda.

  2 comments for “Big data and actuaries – are we part of the solution?

  1. Rainmaker
    May 28, 2014 at 11:22 am

    The solution is not big data but rather bridging the serious disconnect that actuaries have with the rest of the business. There have been a number of occasions where the actuaries come up with analysis of a problem that the management is already aware of…. it’s like basis risk in hedging. Your investment is oil but your buying put options on a gold index

    Like lapses have increased –> yeah we just pissed off a dealer group.

    Mortality increased –> yeah that one advisor who used to write lots of business with us…. he was quite dodgey

    The scariest issue is that more and more senior executives are totally disconnected with the business they are supposed to manage. Lets just write up the FCR, do the valuation, re-calibrate profit margins etc

    As long as this disconnect continues and actuaries fail to understand the stories behind the numbers then whatever you produce will never be appreciated

  2. Charles
    July 12, 2014 at 11:54 pm

    Actuarial, computing and data sciences (and countless others) are all siblings of the same parent, mathematics.

    Data science, when applied towards specific and well-defined business problems, offers benefits such as:
    1. choice of predictive modelling and clustering strategies
    2. selection of learning algorithms appropriate to the task (and data)
    3. reducing features that are correlated and eliminating noise
    4. boosting sparse signals that may be good predictors
    etc. etc.

    That said, actuaries and other insurance professions have accumulated such an overwhelming domain-knowledge monopoly that it is unlikely, that a data scientist, working with traditional actuarial datasets, will uncover novel insights overnight. More often than not, “interesting” findings relate to business decisions or events, not reflected in the data. Without adequate business or domain knowledge, data scientists run a great risk of “not seeing the wood for the trees” and assuming that “historical data is a reliable predictor of the future”.

    Lets pretend that the data scientist is fortunate enough to survive all of these pitfalls and successfully navigates the occasional data cleanliness minefield. This scientist may eventually realise that his/her best chances lie in going cross-disciplinary or cross-domain with the data and begin solving specific business problems such as:
    1. identifying key features or clusters that may improve the accuracy of targeted marketing campaigns;
    2. identifying novel features or combinations of existing features that may improve the accuracy of underwriting assessments;
    3. identifying which policies are likely to lapse (Not sure whether data collected at risk commencement is reliable at predicting lapses based on more recent circumstances? If so, does the reliability deteriorate over time?)
    etc.

    Fortunately for the data scientist (or management consultant), certain insurance portfolios can have such lengthy lifecycles, that the jury may be out long enough, well after all invoices are paid.

    So many “ifs”, assumptions and uncertain outcomes, does not exactly provide business with much confidence. At the end of the day, the data scientist may be more appreciated, by its actuarial siblings, for simple but certain “incremental steps”, such as speeding up risk calculations, experience analyses and other actuarial computations with proven data tools and concepts such as MapReduce (e.g. http://datasys.cs.iit.edu/events/ScienceCloud2013/p01.pdf).

Comments are closed.