The replication game

How ING put together a portfolio replication and economic capital calculation system for its global insurance business within two years. Clive Davidson reports

In October 2005, at an ING internal conference on insurance risk for its global insurance businesses, held in Brussels, Tom Wilson, then chief insurance risk officer of ING, posed the following questions. What would the company do differently in its economic capital reporting if it wanted to get the numbers in five days and report monthly, as opposed to taking 10 weeks to get the numbers and reporting quarterly? And what would the company do differently if it wanted to automate the process of producing the economic capital numbers?

The challenge was mostly met with blank stares, and there were no immediate good suggestions, says Doug Caldwell, manager of corporate insurance risk management (CIRM) at ING. In essence, an economic capital reporting system is a risk model, and the questions required innovative thinking around how to create such a model. Traditionally, insurance risk systems are heavily dependent on actuarial projection models, and these typically can take a whole weekend to run a single scenario to value the liabilities of any reasonable sized business.

In the car heading back to ING headquarters after the conference, Wilson and Caldwell talked the problem over. It was a discussion that would have far-reaching consequences for ING, and would reverberate throughout the industry. In that two-hour drive as they sped through the autumnal countryside towards Amsterdam, Wilson and Caldwell came up with the ideas that less than 18-months later would form the basis of the ING insurance division's pioneering Economic Capital System (Ecaps) - a system that could calculate economic capital on a monthly basis for the entire global insurance business spanning more than 30 countries. The system is highly automated, and more critically, is transparent and auditable - so much so that in its 2007 annual report, ING was able to take the ground-breaking step of disclosing its economic capital for insurance - EUR23.2 billion (EUR36 billion for ING Group as a whole, taking into account diversification).

ING had begun developing an economic capital model back in 1999. It had a head start over many pure insurance companies because it could draw on the expertise of its banking division where this type of modelling was already established. By 2005, the insurance division's economic capital model had matured to the point that the ING board gave the division approval to make it a key part of business decision making, for capital management, product evaluation, risk and return strategy analysis, and so on.

"However, the problem was that the figures (produced by the economic capital model) were not auditable, and it was taking us 10 weeks to produce a report following the end of a quarter," says Caldwell.

The source data for the model - information on liabilities based on a multitude of products issued in a host of different markets, with all kinds of guarantees and options, as well as a whole range of assets - were spread across the local offices globally. This complex array of data fed into a disparate collection of local actuarial systems that were inconsistent across the division, and had variable capabilities and ways of handling data. From there, the risk and finance departments had to collect the results, aggregate them and produce reports. Furthermore, much of the modelling process was done in spreadsheets, which are notoriously difficult to control and audit.

As Wilson and Caldwell pondered how they could action this process differently, they came up with a key insight - if they could replicate the variety and complexity of their liabilities with simplified portfolios of standard financial instruments it would make the whole problem more tractable.

The concept of replicating portfolios had been around in banking for some time for applications such as hedging derivatives, and had been discussed in academic circles for insurance liabilities. The idea is to mimic the cashflows of complex liabilities with those of a simple set of financial instruments; and to use these instruments, which can be more easily valued, to calculate risk. The theory was promising.

"But we needed some sort of industrial or mass-produced replicating portfolio methodology so that we could develop them for all of our liabilities and assets around the globe," says Caldwell. "Technically, we didn't need replicating portfolios for our assets, but we decided to use them on that side as well because we wanted to finish the project in one year rather than the three it would take if we tried to map all of our assets into a new system."

Critical to the success of the approach was to find the most appropriate set of financial instruments to mimic the liability cashflows under a range of scenarios. Enter Antoon Pelsser. As chief quant, Pelsser set about exploring the feasibility of turning the theory of replicating portfolios into practice. The results of his initial investigations were positive, and after around two months CIRM decided to embark on a full scale project to develop an economic capital system based on replicating portfolios.

"The speed with which we came to this decision was highly abnormal in a large insurance organisation," says Caldwell. "Normally, you spend a year investigating, then another year discussing how you are going to do it and who is going to lead it. We committed to the project after two months."

CIRM set themselves the goal of developing the system in a year, which was ambitious given that the insurance division was breaking new ground, and the fact that at the time - February 2006 - it had no internal risk system, nor any available developers. The company needed technology, not only for creating the replicating portfolios, but to deal with the other issues around calculating economic capital: the need to gather and move substantial amounts of data globally, the need to eliminate spreadsheets and for processes to be transparent and auditable, the need for efficiency and automation, the need for consistency, and so on.

For the technology, CIRM turned to two companies - Toronto-based risk management systems specialist Algorithmics, and Amsterdam-based financial technology developer SecondFloor. ING already used Algorithmics' technology for risk management on the banking side, so the company was familiar, and as CIRM was looking to make a quick start on its project, it made sense to look to Algorithmics for the instrument valuation, market risk and capital calculation aspects of Ecaps. However, as it turned out, Algorithmics also had technology that was suitable for creating the replicating portfolios themselves.

"We didn't initially go to Algorithmics for the replicating portfolio capability," says Caldwell. However, after pondering the problem, Algorithmics realised that within its suite of risk tools was optimisation functionality that could be adapted for creating portfolios which replicated insurance liabilities. Caldwell describes the discovery as, "Dusting off some old Algorithmics technology" to fit a new problem.

Curt Burmeister, a director at Algorithmics, says, "The replication problem was essentially an optimisation problem - a matter of choosing a set of assets from a universe of eligible assets to create portfolios that minimise the difference between their cashflows and the liability cashflows from the business units." Although there are a number of optimisation tools on the market (Algorithmics in fact uses a third-party optimisation solver, Cplex from Paris-based Ilog, within its Algo Risk suite), the Algorithmics system was particularly appropriate because it supported a scenario-based approach, and because it could run scenarios on an industrial scale, thereby meeting Wilson and Caldwell's need for a mass-produced replicating portfolio methodology.

With Algorithmics providing the calculation engine, CIRM still needed mechanisms for distributing information to and gathering information from its global business units, translating the information for the Algorithmics engine, storing it and producing reports. For this work, CIRM hired SecondFloor, another technology company with which ING already had a working relationship, and which had a good track record of financial systems development.

ING's replicating portfolio methodology and Ecaps systems works as follows: CIRM uses Algo Risk to generate 500 scenarios that project forward on a quarterly basis for 60 years, with sensitivity to equities, interest rates, foreign exchange rates, and in some cases, inflation. Around 300 of these scenarios are risk neutral, while 200 include high volatility. "We decided during the testing that we would make some of the scenarios regular and some really volatile to make sure we capture all the options and guarantees when they are far out of the money," says Caldwell.

Using the SecondFloor-built intranet, CIRM distributes these scenarios to all its business units. The business units then use their own actuarial system to generate cashflows for their assets and liabilities, including all options and guarantees within them, under the 500 scenarios, and send the cashflows back to Algo Risk via the intranet.

Initially, CIRM selected the universe of candidate assets for the replicating portfolios. So far, these assets have included zero coupon bonds, equity forwards, swaptions, equity options and foreign currency options. However, since they know the nature of their liabilities best, CIRM is in the process of giving the business units the responsibility for choosing which candidate assets to include in the optimisation process. (The Ecaps system includes a 'sandbox' where users can run trial and error replication experiments.) In addition, users can set constraints on the portfolio construction. CIRM has built four constraints into the system - bucketing, value constraints, trade penalty and maximum time to maturity. Of these, trade penalty is the most used.

Using the trade penalty constraint is like throwing sand into the system, says Caldwell. "If you have a linear optimisation program, it's just pure math and the program may be very happy to buy a billion zero coupon two-year bonds and sell 990 million three-year zero coupon bonds, so it can create some really big long and short positions. The trade penalty constrains the program from making additional trades for very small benefit," he says.

Having selected the candidate assets and set any constraints, it is a matter of clicking on go, and the system automatically runs the optimisation program to create the replicating portfolios for the liabilities and assets. "And once you have these (replicating portfolios) you can do all kinds of exciting calculations very quickly," says Caldwell.

First and foremost is to model economic capital. To do this, CIRM generates around 20,000 real world scenarios (Algo Risk already had a real world scenario generator, but Algorithmics and ING jointly developed the risk neutral scenario generator), and pumps its replicating portfolios through these. The real world scenarios include around 500 market and other risk factors, such as interest rates for a number of countries, equity indices, credit spreads and mortality risk. Because the replicating portfolios comprise mostly vanilla instruments, they can be valued using closed form solutions, so the simulations are relatively fast. The output of the scenario simulations is a value-at-risk distribution - or any other risk measure that the company wants.

"Once we replicate (assets and liabilities), the system can spit out any risk measure that we want," says Caldwell. This includes a 99.95% confidence level VaR measure as required by CIRM for economic capital, a capital-at-risk measure as required by ING's board, or the 99.5% confidence level VaR that Solvency II will require.

"We believe that (Ecaps with its replicating portfolios) gives us everything that we will need to apply for internal model approval for Solvency II," says Caldwell. Furthermore, the transparency and auditibility of the system means that it meets the requirements of the International Financial Reporting Standards (IFRS) 7, which requires the disclosure of financial instruments used by a financial institution. "IFRS 7 requires companies to disclose their risk and the way they manage that risk," says Caldwell. "One of the problems ING had was that although we were managing the business based on our economic capital model, the figures were not of sufficient quality to publicly disclose. However, as of 2007 we are disclosing economic capital in our annual report." In addition, Caldwell believes that the system will provide the foundation for compliance under IFRS phase two, which is likely to require market value reporting for insurance liabilities.

One of the other major advantages of Ecaps is that its enables CIRM to spread the work of calculating economic capital more evenly through the year. Because the replicating portfolios automatically adjust themselves with moves in the market, they can be calculated ahead of time but remain valid when calculating capital. "So if we create a model (based on replicating portfolios) as of November month end, we can still use it in December, with perhaps some volume adjustments," says Caldwell.

Another spin-off has been the greatly increased market risk capabilities now available to many of the business units, particularly in emerging markets such as Korea and Taiwan, where they now have an edge in this area over their competitors.

But creating Ecaps and getting economic capital numbers out of it was not without its challenges. One issue, that was met by working with the banking division, was sourcing the considerable amount of market data for the valuation of the various financial instruments in the portfolios. Another that required sophisticated financial engineering skills - and more than a hint of art - was the selection of the universe of eligible assets for the replicating portfolios. This was, at least initially, largely the contribution of Pelsser, now professor of quantitative economics at the University of Amsterdam.

Convincing the business units, for whom the notion of replicating portfolios was totally new and beyond their experience, that the system would work was also a challenge. "Turning insurance products into financial instruments is not the way that insurers are used to thinking about things," says Caldwell. But a concerted programme of education and training eventually won the project support. In addition, user-friendly interfaces for the business units to interact with Ecaps, and which hide the complexity of the Algorithmics calculation engine from the users, have been important for the acceptance and success of the system. SecondFloor have been instrumental in this area of development.

Given the substantial advantages that replicating portfolios give, it seems an obvious route for others to follow. Wilson, now chief risk officer at Germany's Allianz, is - not surprisingly - working on such a project. France's Axa is developing something similar, although it refused to discuss it. Meanwhile, Algorithmics says that it already has three other clients that are now in the process of developing, or planning to develop, replicating portfolios using its system - although it would not identify them other than to say they are major European insurers. In addition, a number of other companies are talking to Algorithmics about the potential for the new methodology in their businesses.

By dusting off and enhancing its optimisation technology, and combining this with its scenario-based market risk capabilities, Algorithmics appears to have caught the traditional vendors of actuarial software on the hop. These vendors, that have long dominated insurance, are now rapidly beefing up their scenario simulation abilities to handle industrial-scale applications being demanded by customers.

Caldwell and Wilson's two-hour discussion as they motored from Brussels to Amsterdam must rank as one of the most fruitful travel conversations in recent insurance history. The results of the ideas they sketched out then, barely two and a half years ago, are available for all to see in the disclosed economic capital of ING's 2007 annual report.