Session 5106: Impact of MCCSR Changes on Pricing Session 5106 : Incidences des changements du MMPRCE sur la tarification

Transcription

1 Impact of MCCSR Changes on Pricing (Session 5106) 1 Session 5106: Impact of MCCSR Changes on Pricing Session 5106 : Incidences des changements du MMPRCE sur la tarification November 11 Novembre 2005 Moderator/Modérateur: Speakers/Conférenciers: Marc-André Belzil (Brian K. Jenkins) Marie-Josée Blanchet David C. Gilliland Moderator Brian K. Jenkins: Good morning everybody. You may notice in the program that the Moderator of this session is Marc-André Belzil. Today, I am Marc-André Belzil, since unfortunately he can t be with us today. I checked, and Marc-André was one of the people on the sub-committee that did research in the development of the new Minimum Continuing Capital Surplus Requirements (MCCSR) formula that we re presenting today. And fortunately I noticed that there are at least one or two other members of that committee here, so it won t be that Marc-André was afraid to come up and be participating here and see that. The goal of this session today is to provide practical views on how the new mortality component of MCCSR may impact on pricing. We will look both at the factors themselves, as well as the mechanics of how this component will perhaps be considered in pricing. I hope that will give you a good overview. It s a working session. The presenters encourage you to ask questions as they go through their presentations. You can wait until the end to ask questions as well but they would encourage you to ask questions as we go through. As I ve mentioned briefly, the new formula was created partly out of some research that an MCCSR review sub-committee worked on. Today, I think Gord is here and he was on that sub-committee, and Ben Meckler was the chair of that committee. My observation is that there was primarily a focus of financial management or corporate people more so than pricing people in the development of the formula. And that s not a bad thing but pricing people should also be paying attention to things that can affect their modelling and their work. And I would encourage pricing people to get involved in CIA committees and subcommittees to help make sure that their perspective is also considered in putting things together, so that we ve got a broader perspective. So you can t rely just on the financial people to do all our work for us all the time. We have two wonderful presenters today, very enthusiastic. A comment from one of them today was, after this morning s session with the broad view of the Canadian perspective and everything else, we re now going to get into the true Actuarial post, a real tunnel vision on one specific component of mortality, MCCSR on pricing. Our first presenter of this morning will be Marie-Josée Blanchet. Marie-Josée works at RGA in a pricing role. She does a lot of analysis internally with our pricing models. She s been looking at this issue with the new mortality factors with our Appointed Actuary and his group and with Marc-André as well. She s going to bring a very good perspective to this.

2 Incidences des changements du MMPRCE sur la tarification (Session 5106) 2 Our second presenter is David Gilliland, he works at some company that does pricing models or something, and I think it s called GGY. I m not sure many of you understand or have made use of the AXIS pricing and modelling system. He s been involved in CIA activities; he s been involved in investment practice, the Organizing Committee for stochastic modelling, which I think we might need to do some of the volatility analysis properly on this. He s currently involved with the ALM Working Group and is the CIA representative on the Enterprise Risk Management process task force. So, we start with Marie-Josée and please don t feel shy about asking questions. Panellist Marie-Josée Blanchet: Good start. Okay, first we re going to look through the evolution of the MCCSR mortality component and then afterwards we will just compare the component that was in place pre-2005 to now. Then we ll see what it means for pricing, and we ll look at the issues and some observations that we ve made looking at the new formula. And then I m going to talk a little bit about a survey that I ve done with life insurance companies, in which I ve asked them where they re at in the implementation of the new mortality component. You will see that on the corporate side it seems that companies are pretty much there; they ve estimated it. But on the pricing side we re not there yet. I ll show you the numbers a little later. The old formula was implemented in There were factors varying from.005 to.002 and it was adjusted for net amount of risk and for a statistical fluctuation factor. There are several reasons why we needed to change and update the prior formula; one of them was mortality improvement. Since that time there s been quite a bit of improvement, there s also been a shift in insurance population. There has also been an evolution of products. Before there was no recognition of product profile, standard deviation, or duration; it was just a flat two dollars per And also the reinsurance risk was not adequately reflected. If we look at the reserves right now, they provide for the permanent deterioration, and also for mis-estimation of the mean. The capital is there for volatility and also partly catastrophe. The new requirement is the sum of two components; the first component is for the volatility. It s computed for each set of life products. There s a part of this formula which is the standard deviation, another part which is the log normal of the duration and another component, which is net amount of risk divided by net face. When we talk about a set of life products, the definition is all products within a set should have similar attributes with respect to adjustability and mortality guaranteed duration. In the volatility component, the standard deviation is the product of qx, (1 q) and b2, where b2 is the net face amount. The q here is using the valuation mortality, which includes the Provision for Adverse Deviation (PfAD). The factor B is the natural logarithm of the duration of projected net death claims using a 5% discount rate. And if it s an adjustable product, you can also use 50 percent of this factor. For group business, there s a different formula. We re not going to talk about it, because this is an individual session and I haven t worked with it, so let s move on. After we have calculated the volatility for sets of life products, we calculate the overall volatility for the company just by summing the squared volatilities and then taking the square root. The second component of the requirement is a catastrophe component, which is 0.1 times the next year s death claims and then there s also a 50 percent factor for adjustable products. The catastrophe component is added to the volatility to get the overall mortality component. OSFI believes that there s good reason to implement this new formula; it s going to be starting at year-end 2005 and fully implemented in third quarter of Délibérations de l Institut canadien des actuaires, Vol. XXXVII, n o 1, novembre 2005

3 Impact of MCCSR Changes on Pricing (Session 5106) 3 So if we look at what it was before and what it is now, and if we look at the catastrophe component which is the easiest, then you just compare two dollars per 1000 to 10 percent of claims. So you see at the beginning, the new component is much lower but then with time it goes higher. For the volatility component, I cannot have a graph like that because it depends on the company, and the products. So in the past when we ve been pricing, we ve been looking at per policy pricing, we were making sure that expenses were all allocated and also there was part for MCCSR components and so on. Everything was allocated; you can see what the price is for each policy per se. With this new component, can we do this? If we look at the component, then it seems that at a point in time, it s fairly easy to implement. Those are all values that we have so for corporate, I think, that s the reason why they have already started to go towards and implement the new methodology. You know the next year s death claims, we have the information in the system to be able to get that. Also for the other components, it might be a little bit more difficult but the information is in the system and you can modify somewhat your system and be able to get that information. So as I said it is pretty much information that we have. The duration part, we already have some information with Dynamic Capital Adequacy Testing (DCAT) and other projections and we can also get that. But for pricing, we need to project that measure many years in the future. You cannot price just one product independently of others, because you have to calculate the volatility component at the company level and then allocate it back to the policies. So I ve done just a really small block of business, new business, whole life and I said, okay, what s the volatility for 1000 policies and then increase the number of policies and see what it meant for volatility. Well as you can see, it varies a lot by the number of policies. Dave will discuss much more; he did some sensitivity of results depending on different number of policies and so on. So I think it s going to give us a good idea of how the catastrophe and volatility components move with time and different assumptions. So as I said, we cannot price policy by policy. So to get the volatility component for several years, how would you project that? Well the theoretical way to the exact number would be to look at your inforce, project your new business for the next 100 years and also you have to project all the other business- not only the product you re pricing or the set of products - you also have to do it for all the products of your company. Also, the way we ve been pricing for years is that we use an average policy, maybe with a different bend, but an average policy. Well does that work with volatility? And the second issue is once you have that component projected for the company, how are you going to allocate it back to the policy? So from now on, we re going to be discussing how we should take those issues and deal with them. So if anybody has suggestions or questions, you can just stop me any time and we can discuss. Well one of the solutions may be we can use DCAT work and see how volatility moves with time, but of course this is only for five years. The other thing that we can do is to work on understanding our business. I think that if you do multiple scenarios, you get to feel better about your business and how the component moves with different assumptions just like what we are going to see in Dave s results. I think we re going to feel better and say, okay, I see my volatility is pretty much level or I expect that my business is going to go up in the future and I know how the volatility component is going to move. And so this way we can tell what type of volatility to assume for the future years. For some of you, your company is very mature and you don t expect any changes. So once the volatility is calculated at this point of time, then maybe you can say this is it, I m going to use this forever. Of course, there s inflation and there s different things, but that s one way to see it. So modelling the whole business, past issues and future years, and the whole company altogether, I think it s not possible. So what we ll need to do is to go in the software and by-pass the actual calculation and say okay, here is what we expect for volatility and how we re going to allocate it back to each policy.

4 Incidences des changements du MMPRCE sur la tarification (Session 5106) 4 What is the solution? Let s say I calculate the volatility of the whole company and I want to allocate it 100 percent back to those policies that I have, well how am I going to do this? What if I calculate the volatility of each policy and I divide it by the sum of all volatilities of every single policy. Well, this formula is mathematically correct, all my volatility is going to be sent back to all policies and I m going to be covered. Is this the right way to go? First of all is it feasible to calculate such a thing, are we able to calculate volatility on each policy and then add it up to the company level? Is the ratio a little bit meaningless? Imagine what the denominator of this is? It s quite high, so you re going to get well, you know big numbers times a smaller number, and I don t know. Is this going to be too volatile changing with years? That might be a possibility but there s also some negative to this. Another way to try to allocate this back, is to use a factor per 1000 just like the two dollars per 1000, but we can adjust it to match the total volatility that we have for the company. One way would be, you calculate the volatility you have for the whole company divide by your net face or by your amount at risk and then you already have your factor per Now doing this, is it valid? Well maybe, but there s a part which is linked to duration, and for term plans, probably the volatility should be lower than for whole life. But what we could do also is look at the volatility of let s say, term insurance. Look how much it gives to the whole company and allocate it back, so you have a different per 1000 rate for permanent versus term versus other type of products. So that could be a way, again it s not the exact contribution to the overall portfolio because I did not account for the ages. Volatility varies by age and so on but we ve lived for few years with a flat two dollars per 1000, so maybe we are able to price a product using little bits of subsidies between ages, as long as we know that on the whole picture, we have allocated back the entire mortality component. I think this is an easy way to do this, so we re going to ask corporate and they can give us the answer on how much it would be per 1000 for each set of products. So I did a survey with a number of companies to see where they were. I surveyed 11, well actually 12, but I think there is also a good number here. First question: were they aware of the change to the mortality component? Well, good news, yes. Second question: Have they estimated it at the company level? another good thing - yes, they have. All companies have actually estimated it at the company wide level. The next question I asked is, are you using it in your Dynamic Capital Adequacy Testing (DCAT)? Well, many companies are probably waiting a little longer to complete their DCAT, so it explains the fact that only a couple of companies have actually used it for DCAT. And my last question is have you used it in pricing? And now that we ve seen the implementation problems, I think this is a reason why only one company actually has used it in their pricing. I think that in conclusion that we have to change the way we are doing pricing for that component. We have to decide on how we are going to do this at the company-wide level and then how we are going to allocate it back. I think the first step is getting to know your business. What happens with this component? What do you expect in the future? And do some tests on the number of policies, what if you have a little bit more business, what if there s inflation, how is it going to affect your volatility? Quite a few of those tests have been done by Dave and he s going to show us his results and also I imagine there s other ways to do this. Maybe allocating per 1000 is not the right way to go? What I ve heard is some people have been looking at per thousand rates and also some people are thinking of a per volatility component, like a percentage of the volatility that the software would give you. And some of the people said, well may be I would go with adjusting the component for catastrophe and put it there like a percent of premium of some sort. That could be another way to go. So I think we need to do some testing and decide on what type of factor and how to implement this. Délibérations de l Institut canadien des actuaires, Vol. XXXVII, n o 1, novembre 2005

5 Impact of MCCSR Changes on Pricing (Session 5106) 5 Panellist David C. Gilliland: Good morning. I d like to thank Marc-André and Marie-Josée for asking me to join the panel to discuss implementing the new mortality MCCSR components in an actuarial model. I ll describe how we incorporated these changes into AXIS; however I think most of the concepts are not specific to AXIS and Brian, there will not be any stochastic modelling in this presentation. I ll give you some background on the AXIS structure as it relates to these calculations. AXIS is the cash flow modelling system used by many Canadian companies. It is used for different purposes, so as was mentioned, the problems of pricing and projections are different for this particular component. It includes a number of different approaches to required surplus and it is based on a hierarchal model structure that allows combining model points at a higher level. This is a picture of that hierarchy. How do you build a model of your company in our system? The model s built up from the lower cells; the results are calculated at the cells and then passed up to objects we call funds, sub-funds, funds and offices to build a model of your whole company. The funds might be for example your par fund and your non-par fund, that type of thing. Most of the calculations are actually done at the cell level and results are summed at the cells, stored at the cell, and then passed up at a higher level. So one of the things that we like in a model like this is for things to be linear. Anything that s linear is great; you calculate it at the cell, and you just add it up as you go up through the model. What we don t like is non-linear. So one of the things that we can do at the higher levels is calculations in aggregate that are not linear and can t be added up; that s one of the purposes in our model for the higher levels. For example in the required surplus there s an aggregate cash value deficiency calculation; it can t be calculated one model point at a time and passed up. Okay, so what I m going to talk about is how this component can be modelled as Marie-Josée said. She described the new mortality component. It has two pieces and I ll spend most of the time talking about the volatility piece but I just want to point out that the catastrophe component as you see here is basically a component using the expected claims. So you would think this is a good thing; expected claims are linear. You can add them up in that structure as I was discussing. You would calculate expected claims at lower level and add it up and so this component should be linear. But it has a term, the last term E over F, that s calculated at a set level in the requirement, which means that in fact it s not linear. So while you might think you can calculate this at the lower level and add it up, when we think of some simple examples, you can see that that s not the case. So even the simple component isn t quite linear, although it s much more linear than this other thing which is one of the formulas in the new volatility component. In this calculation, there are two problems. Since I don t like it to be non-linear, one of the things I don t like to see are square root signs in front of us. The component A is the square root of the sum of the squares of the volatility components for the various sets of products. Then for each set of products the formula is A times B times E over F, and again we have the E over F factor which is not linear, but in this case the A also has a square root of the sum of something. The sum is of the variances of the individual model points or policies. Again, this is a problem for doing linear calculations and passing them up to a higher level and I ll talk about that a little bit. The rest of the formula, the natural logarithm of the duration of your set of products again is not linear, that s bad. So when we first looked at this formula we thought, oh good a standard deviation. We already have a standard deviation calculation in our system, so we should be able to use that. I think Marie-Josée said in her remarks that these things are easy to calculate; unfortunately they re not that easy all the time. The standard deviation calculation we had in our system prior to this was to model experience refund calculations where it was a standard deviation to be used in a stochastic experience refund calculation. It was based on the pricing expectation and it was also based on the number of lives starting the model. So what s the difference between that and what was required for this? This is a standard deviation at each point in time as you move forward in the projections. It assumes that you have reached that point in time and then you want to do a projection based on your valuation assumptions instead of your pricing assumptions. So there were two reasons we couldn t use our existing calculation for factor A.

6 Incidences des changements du MMPRCE sur la tarification (Session 5106) 6 Then we saw a duration. We had a duration calculation in the system too but that wasn t the right one either. They ve decided to use a different definition of duration - a Macaulay duration - on only the expected death benefits, not on the whole policy, and discounted at a flat five percent regardless of the rest of the assumptions in your model. So we had to develop another calculation for that. There were a couple of values in Axis that we could use: both the net amount at risk calculation, E, and the net face amount we re already included in our projections, so those values were available. The new lines we added to our calculations of what we call calendar year results were the Macaulay duration lines and, as I mentioned, this isn t linear, but we can calculate this at higher levels by summing the numerator and the denominator, which are linear. So what we do is we calculate at the lower level the numerator and denominator for the duration calculation, pass those numbers up to the higher levels, and then just divide one by the other to get the duration. The expected claims, C, which is linear, we already had. We added a line for variance, which is A squared rather than A in the formula, because the variance is something that you can add up, but you can t add up a standard deviation. The volatility component squared is the thing we store, as is the catastrophe component. This picture is meant to represent the cell level in AXIS and reason it s in here is to point out that there s a couple of ways to run a cell. One is the seriatim policy-by-policy calculation and another way is to run a model using model points in a volumes table. And the distinction there is that when you are running model points, usually one model point will represent a number of policies instead of just one policy. Again if things are linear that s not a problem but for things that aren t linear you have to think about the implications of that. Usually for each model point we use one average size and one average size for a group of policies is okay, again, if things are linear. But if they re not, you need to think about the impact on your required surplus of using one model point to represent a number of policies. I think the other point about the cellas I ve already mentioned is that the values are stored at this level. This is the only level in our system that we store results for and when we run the higher levels we just add up all these stored results from this lower level. This is another picture of the hierarchy just to mention that some of the calculations vary by modules and these new calculations impacted the life modules that we call Regular Life, Universal Life and Par. The objective of adding all those new lines was to do the correct calculations of the components on a total portfolio and so we built a model. Now we don t have our own block of business to model, so we just made one up. So it may not make much sense to one company but it s just to give you an idea of the impact of these calculations and what they might look like. We only included two plans, whole life and term and a spread of issue ages and issue years, assuming that the issues were increasing five percent a year and all the face amounts are $100,000 with no reinsurance. The in-force at the start of this model is about $25 billion so you can relate that to your company and this is what the components look like. It looks like we are into Christmas colours already. The catastrophe component you can see from this is flatter; this is the component expressed per 1000 net amount at risk for the block I described of in force business. There s no increase in business over time, so what you can see is going to happen to you. If you try to model your business and assume there s no new business in the model, run out your model here over something like 50 years, the component is going to get very large. And here s another projection if we include new business so this makes a big difference. So the thing to take from these two pictures is that you re trying to figure out an average factor since as Marie-Josée explained, in order to price this you need to figure out some way to allocate these aggregate level calculations down to the lower level. The average factor can vary quite a bit, depending on your assumptions. Going back to this picture, the number in the middle of that scale is something like 20 per 1000, and this scale is more like one per So there s a lot of variability depending on well how you build up your aggregate model. Délibérations de l Institut canadien des actuaires, Vol. XXXVII, n o 1, novembre 2005

7 Impact of MCCSR Changes on Pricing (Session 5106) 7 Getting back to the question, how to do the calculations at the lower level? I ve probably already mentioned the words linear and non-linear enough. If you just take the simple case, let s look at the formula for one policy and then assume that everything in this calculation is the same except the number of policies. So here s the formula for N policies. And so you see the only thing that changed is that inside the square root sign there s now an N, so we go back to this first slide and if we actually calculated N policies individually and added this up, assuming it was linear, the N would be on the outside of the square root sign and on this slide, it s on the inside of the square root sign. And so, if we have N identical policies, by very simple algebra the total component that you will actually get is one over root N of the component you would get from just adding up the components from the individual policies. So there s a very big size factor here and that s apparent when you run the model. Once we put this new calculation in the system, one of the runs we did is running this with one policy and here s an example of that, and the components you can see on this slide are components per So that was kind of worrying; we thought we had the calculation wrong but you can work out the calculation for your own examples. These components are very large capital requirements and you certainly wouldn t want to do any pricing with these components, calculated policy by policy. Here s another example with policies and you can see that the graph is exactly the same except the scale has changed because it s exactly the same policies. The scale for policies has changed by the square root of You ll see that here where the scale is in thousands, the scale has gone down and the top number is 90 instead of 9000 per thousand so it s gone down by a factor of 100. This volatility factor is very sensitive to portfolio size. What this next slide shows in the middle column is the scale factors from the old MCCSR mortality formula. If the scale factor you get at around ten million dollars of component (that s not volume, that s the total mortality component) was 1, then the lowest factor for a larger company would be 0.6. This third column is just estimating what the factor would be using that one over root N approximation; instead of 0.6, we have 0.1, just reinforcing again this idea of the greater variation by portfolio size. We ve said a number of times the volatility is not the sum of the volatility component calculated at the lower levels. It s nonlinear, so how can this be allocated to lower levels? Well, there s different ways to do this using factors; we investigated what the factors would look like per face amount and per 1000 net amount at risk and another way would be as Marie-Josée suggested. That is, sum up the volatility factors calculated at the lower level and then recalculate the volatility at the higher level and take the ratio of those two things. This slide shows you for that in force block that I showed you earlier, what the factors would be if we used factors to allocate. The red line is per 1000 of net face amount, the green line is per 1000 of net amount of risk, and the blue line is the factor if we use the sum of the volatility components and used a ratio to that sum of the volatility components. Now this is the same picture, but we ve added level new business and here is the same thing again if we had increasing new business. So again the point is for any of these methods of allocating, you can see going back over these different slides that the factor is going to vary a lot depending on the model of your total business. What s the impact on pricing of these different methods of allocation? I mentioned this point already that if you use the average size in B in the formula instead of distribution of sizes that would change the variance term. And if you think about using a per 1000 factor versus using a proportion of the calculated value in the cell, that could have a very big impact on the ROI calculation. If this is the shape of the volatility component calculated in the cell and you use a proportion of this in your pricing as your capital requirement, that s quite different than if you re using net face amount or net amount at risk. The net amount of risk line compared to these two would be a straight line. So if the capital requirement is expressed as a straight line as opposed to something that increases a lot over time that s going to have a very big impact on your ROI calculation, if ROI is one of the things you use for pricing.

8 Incidences des changements du MMPRCE sur la tarification (Session 5106) 8 Another thing to think about in using a factor approach like this is the marginal pricing issue. What s the marginal impact of writing some amount of new business? For doing your pricing, you could translate the total component as we mentioned into one of these factors. If you then add new business to your model, you ll get a new factor. Then for your pricing you must decide should you use the average factor for your new model, the factor from your old model or the marginal factor for your new model. I m not sure why, but I think because of some simplifying assumptions, this projection is a bit less smooth than I would like to see, but what we ran here shows adding the same amount of new business to the model with level new business. Remember the model started out with 25 billion in force, so this was adding 10 percent to your new business in one year, and figuring out how much that impacts your capital. The total increase in capital that we re comparing here is for two blocks of business, and the only difference in the two blocks of business is the average size. One of them was a $100,000 policy size just like the other one and the second block of business the average size was changed to $1,000. While the total amount of business you ve written is exactly the same, that change has a huge impact on the amount of capital that you re going to need and that s what this picture is trying to show you. And so one of the things you have to think about is, for example, if you are reinsured and you are writing large blocks of business. So you have to think about this marginal cost but you don t necessarily want to just place emphasis on the marginal cost. I think, in this example the factors per 1000 for the marginal costs were both below the average factor. So if the average factor for the block of business that we started with was something like 90 cents per 1000 required capital at the start of the projections and after we added the new business in it didn t change that much because the new business represented only 10 percent of the business. But the marginal factor would have been something like 85 cents for the top case that assumed 100,000 dollar policies and fifteen cents for the hundred 1000 dollar policies. So it s not clear how to reflect that in the pricing. If you make a decision that you are going to use an average factor, then neither of these blocks of business would be affected. The pricing would be affected if you consider the relative change in the absolute value of the required surplus that you need. So people are going to have to think about how to factor that into the pricing as well. Conclusion. You need to model your business, figure out what the impact is on your company. There are a number of issues in setting up the model and that will have an impact on the result and one of them I mentioned was the average size. So, it is a problem but it s not a new problem; we do have size adjustments already in Required Capital, and companies have ways to deal with that. We also have size adjustments in terms of the marginal pricing issue in things like expenses. You just need to develop a corporate philosophy on how you are going to handle this; probably a factor-based approach will be essential to do your pricing. And you need to investigate and determine what s best for you. Thank you. Moderator Brian K. Jenkins: Thanks Dave and Marie-Josée. I was prepared to come here and participate in the session and ask questions but I got to be the moderator. So before I even get to ask questions, I wanted to give you five observations that I have with respect to this formula and I m sure it will wake a few people up. My general view is change is always an opportunity. I view pricing actuaries as innovative and creative people; they have the ability to dig into and understand all these formulas and how they work together, to create significant new opportunities in what they do. So I would encourage the pricing actuaries who may have skimmed over this note or said, oh yeah I ve heard about the change to really look at this in depth and understand it because I think if you work at it you ll really start to get some good understanding and come up with some ideas. The second observation was that, if I was a pricing actuary and my boss maybe had a physics background or something, what a great creative discussion we would have about how we should take advantage of this to create pricing. So find out if you ve got a really strong technical boss, and I m sure some of you pricing actuaries do, but talk to people who have that background Délibérations de l Institut canadien des actuaires, Vol. XXXVII, n o 1, novembre 2005

9 Impact of MCCSR Changes on Pricing (Session 5106) 9 and see what opportunity you can get from them because these formulas are pretty basic, but when you put them all together and work, they can show something interesting. I observed that the group approach says, well, there won t be enough data so use this approximation. Now take a look at the group piece, think of the individual piece, and look at how they add together. There might be an opportunity in looking at both group and individual together, and you might not want to be giving in an aggregate sense the marginal benefit of improved capital to the group side since you can lose it on the individual side. So you want to make sure that you re taking advantage of that on the individual side as well. Group wasn t talked specifically about here but it seems to have been thought of afterwards, oh we don t have the data, so let s use an approximation. Approximations can sometimes create even more advantage but just be aware, the Office of the Superintendent of Financial Institutions (OSFI) might be thinking about the fact that if you take advantage of it inappropriately, they ll have to fix that, so use that appropriately. Now come to think of it maybe two of the most significant observations I have with respect to this new formula came primarily from financial management people, not pricing people. And if you noticed in Dave s presentation just running off in force has a huge capital increase requirement, which means that to model this stuff right you have to model in force and your new business together. Now, who does that? Pricing actuaries don t usually do that, do they? Financial guys do that. So, maybe we don t need pricing actuaries anymore and this is kind of a secret way to get rid of the pricing actuaries because now the modeling will all be done using DCAT in an aggregate sense and in some cases, people will believe that will bring rationality to the pricing of new business. (Audience laughing) The other observation was that right now in the current market place, the MCCSR factor for mortality is very clear; it s marginal, and you can add it on. So the reinsurers spend all of their time talking to whom? They talk to the pricing people because they re working on the marginal increase in capital; they re looking at mortality. Well now when we get into this complicated formula, who are reinsurers going to have to talk to about the optimization of MCCSR with regard to mortality? I think the conversations might change from the pricing actuary to the financial management group, to the Appointed Actuaries. I think it s another way for the financial and Appointed Actuaries to get more time with reinsurers, so those are my key observations. We do have a good amount of time; the presentations had anticipated that there would be good participation from the audience, so I d really encourage you to share some of your own observations in thinking about this or if you have some questions, please don t be shy. Audience Speaker U-M:?? With all the consolidation we had all ready this?? ( 48:40). Moderator Brian K. Jenkins: That could happen yes. (Audience Laughing) Audience Speaker U-M: The other thing that occurred to me is that instead of selling 1,000,000 dollar policies, I ll start selling 10 $1,000 or $100,000 policies to every policyholder Panellist Marie-Josée Blanchet: Can I suggest reinsurance? (Audience Laughing) Moderator Brian K. Jenkins: But John, and I mean to be extreme on this, if you look at the group formula, if you don t have a big component of group, maybe you want to pick up a tranche of group business from somewhere that s homogenous and

10 Incidences des changements du MMPRCE sur la tarification (Session 5106) 10 benefits from the number of lives to create a significantly lower volatility. So there s ways to optimize that volatility factor that maybe weren t anticipated with the new formula. Audience Speaker U-M:?? (no mike) (49:44 to 49:51) Moderator Brian K. Jenkins: I can t answer that exactly but I would anticipate that there could be different answers, because of the way the formula s going to work. Different people working at things, looking at different perspectives are going to have different views around what the value is. It could be that smaller companies could be a phenomenally attractive acquisition target now because of the impact of their business on the acquiring company, and the smaller company might not actually have the awareness of that when they re being targeted. I m sure Gord has something that he can provide to us, the theoretical background of why this is great. Audience Speaker Gord: We re just starting to look at this from our own company point of view and this raised some very good questions which I don t have the answers for. I did have a question for David. I noticed in your modelling you assumed one average size and I was a bit surprised at that given that the variability in the sizes creates part of the volatility. So I wondered if you had done some modeling with a model office of different sizes as well as different age groups to see what the impact would be. Panellist David C. Gilliland: Well, as we said, we don t have our own block of business to model so we needed to set up a simple model just for illustrative purposes. The point of the last slide comparing a new block of business with a 100,000 average size versus a 1,000,000 average size was to show that it s very sensitive. It is very sensitive to that assumption and I think it s also very sensitive to the distribution of average size in your in-force block and that was again why I made the point that even when people set up models, they usually use an average size for some of the blocks. We have a distribution by age, for example, and you just use different averages by age in determining each of those average sizes. There s an age distribution assumed, and it depends on that distribution. We did a quick test and for a small variance in the average size it didn t make that much difference in the factor, 10 percent type of level for a small variance but we don t really have a block to compare, a realistic block, so we decided not to try to model that for today. Audience Speaker Gord: And my other point or question is on the type of reinsurance. If a company s reinsuring the block, my question is whether you use quota share type of reinsurance or excess reinsurance over certain retention limit? This has a big impact on your volatility component and that may change how companies do their reinsurance to take advantage of those opportunities. Panellist Marie-Josée: I think it s a good observation. I think you want to re-look at how you do reinsurance if you do excess. Of course, it s going to have a different impact than if you use quota share. Audience Speaker U-F: Is there a lapse assumption for what you did David? And what sort of lapse assumption was it? Panellist David C. Gilliland: The policies used for these calculations were sample cells that we have in our datasets to show people basically how the system works rather than actual policies; so they have an example of whole life lapse rates in them. The other cells were term, so some of the products had term lapse rates and some had whole life lapse rates. Audience Speaker Ralph Ovsec: Dave and Marie-Josée, you both talked about capital projections for pricing and whatever you decide internally, it s going to affect you only internally. Have you thought about what public companies who disclose embedded values might want to do or maybe should do because you re sending information out to the analysts? The run off of this mortality component under the new formula is going to be very, very different than what it is under the existing formula and I m sure that under the existing formula people are just saying, well, I suppose it s two dollars per 1000 times my statistical fluctuation factor. Let s say you are Manulife and your factor is 0.6, you just run that off forever. Délibérations de l Institut canadien des actuaires, Vol. XXXVII, n o 1, novembre 2005

11 Impact of MCCSR Changes on Pricing (Session 5106) 11 And the factor moves so little, that s going to have very, very little impact. But under the new formula Dave, when you had the one where your in-force increases by factor of 100, your component goes down by a factor of 10 as well cause you went down from a dollar to 0.1 but that 0.1 is going to run off. The run off is going to be much steeper. So back to the question, what do you think, either of you, the direct companies who are disclosing embedded values should be doing? Panellist Marie-Josée Blanchet: Well, I think that when you calculate embedded value of just in-force business; you want to know, okay, what is the value of the company as a whole? Unless you are thinking of stopping new sales, I think if they just calculate an overall volatility component without new business, I don t think it s correct. I think it s going to be projecting your way to high capital. Panellist David C. Gilliland: I think that was the point on that last slide, to think about do you have a going concern philosophy? Presumably most people do for your company and so you model it in that way. But if you have a run off situation, then the capital becomes more and more expensive as the business evolves over time. So that may mean it s more advantageous to sell that business sooner. Moderator Brian K. Jenkins: I m curious Ralph, you asked the question, what you would think would be appropriate in how the embedded value should be determined? Audience Speaker Ralph: Well, these companies are disclosing publicly and when they disclose embedded values, typically there isn t any franchise value. It s not prepared from that standpoint so maybe not disclosing it on a going concern basis. There is no future new business value that you re putting in, although I agree with what the panellists are saying. You ve got to factor something in because it s not realistic that your capital is going to increase steeply because at least for the next year, you re going to be writing new business. I m not aware of anybody who s in such dire straits that they ll be stopping writing new business. The important issue though is going to be disclosure. I don t think you can just say, well, I calculated using the new rules. You are making an adjustment to the formula by factoring in some anticipated levels of new business and the impact that s going to have on your overall capital requirements. Also when you disclose the new business, I m not sure how the cost to your capital is going to be reflected. Are you on a consistent basis with the way you are projecting your overall capital or are you on a marginal approach? Currently, I think under the existing formula, you are inherently on a marginal approach because you are using, let s say, the Manulife s of the world 60 percent statistical fluctuation factor. So they are taking advantage of the impact that the existing business has and the run off is so small in that statistical fluctuation factor that I don t think is significant but we haven t examined what the impact of the two different slopes are because we re grappling ourselves with what the runoff is going to look like. Moderator Brian K. Jenkins: My intuition tells me that if we can come up with the answer to Ralph s question, it will greatly help answer how to price your new business as well and then you can disclose it internally, so that they can disclose it externally. Are there no more questions? I want you to raise your hands how many people are going to run off, get that new mortality component piece and really dig into it, roll up your sleeves and understand it. (Audience laughing) I do encourage you to take a look at this in more depth as I think it does have some interesting opportunities for everybody. Thank you very much for your attention to this. (Audience applauding)