Nova Scotia. Private Passenger Vehicles. Oliver Wyman Selected Loss Trend Rates. Based on Industry Data Through December 31, 2016.

Transcription

1 Nova Scotia Private Passenger Vehicles Oliver Wyman Selected Loss Trend Rates Based on Industry Data Through December 31, 216 Selected Trend Rates Summary The following table presents our selected past and future annual loss cost trend rates as of December 31, 216. We discuss and present our methodology and assumptions in selecting our trend rates in this report. Coverage Past Future Bodily Injury +3.% +3.% Property Damage & DCPD +3.% +9.%* AB Disability Income +.% +.% AB Medical/Rehab +5.% +5.% AB Funeral 9.% 9.% AB Death 5.% 5.% ABTotal +3. % +3.% Collision.5% +4.5%* Comprehensive +7.% +7.% Specified Perils +7.% +7.% All Perils +3.5% +3.5% Underinsured Motorist +5.% +5.% Uninsured Auto +9.% +9.% *Beginning April 1, 213 Introduction Loss trend rates are factors that are used to determine rate level indications. They are applied to the experience period incurred losses to adjust for the cost levels that are anticipated during the policy period covered under the proposed rate program. The application of trend rates is, essentially, a twostep process. The data in the experience period under consideration must be adjusted to reflect changes in cost conditions that have taken 1

2 place (i.e., past trend ), and then the data must be further adjusted to reflect changes in cost conditions that are expected to take place between the present time and the time during which the new premiums will be in effect (i.e., future trend ). Therefore, past trend rates should reflect the underlying trend patterns that occurred during the experience period, which we have assumed to be the three to five years ending December 31, 216. Future trend rates should reflect those same patterns that occurred during the experience period, as well as the likelihood that those patterns may change. We select trend rates based on historical Industry Nova Scotia claim experience as published by the General Insurance Statistical Agency (GISA). The Industry data is organized by halfyear, and in this report we refer to the first half of an accident half year as XXXX1 or XXXX.1 and the second half of the accident year as XXXX2 or XXXX.2. So, for example, the accident halfyear spanning July 1, 216 through December 31, 216 is referred to as 2162 or We derive indicated annual loss trend rates based on an exponential regression model using Industry historical accident year loss and loss adjustment expense data that we project to ultimate cost level (when all claims are reported and settled) using the Industry loss development factors we select. Estimation of Industry Loss and Claim Amounts The Industry Nova Scotia experience upon which the loss trend rates are based must be adjusted to an ultimate claim count and loss amount level. We do so through the application of what are referred to as development factors to the reported claim counts and claim amounts as of June 3, 216. We select development factors based on a review of the Industry Nova Scotia loss development patterns; we do this by coverage. Our selected development factors are generally based on: (a) the volume weighted average of the last four observed development factors for the halfyears ending June (for development period 6 months to 12 months) if there is evidence of seasonality 1 ; and (b) the volume weighted average of the last six observed development factors (for the development periods beyond 12 months). For the more minor coverages, such as Specified Perils, Death Benefits and Funeral, we tend to select the volume weighted average of the last twenty observed development factors through 66 months of development and the allyear volume weighted average for development periods beyond 66 months. For Uninsured Auto and 1 Evidence of seasonality was found for Bodily Injury counts, and Comprehensive counts and amounts. 2

3 Underinsured Motorist, we tend to select the allyear volume weighted average for all development periods. The exceptions are as follows: Bodily Injury Claim Count 612 average of last three semester values Bodily Injury Claim Amount 612 average of last three semester values Property Damage Claim Count 612 Weighted average of last six values Property Damage Claim Amount 612; 96114, Weighted average of last six values; 1. Benefits Medical Claim Count 612 average of last three semester values Benefits Medical Claim Amount 612; 336; Weighted average of last six values; Weighted average of last ten semester values; 1. Benefits Disability Income Claim Count 612 Weighted average of last six semester values Benefits Disability Income Claim Amount 612; 132+ Average of last six semester values; 1. Uninsured Claim Count Uninsured Claim Amount Collision Claim Count 612 Average of last three semester values Collision Claim Amount 612; 48+ Average of last three semester values; 1. Specified Perils Claim Count Specified Perils Claim Amount All Perils Claim Count 612 Weighted average of last six semester values All Perils Claim Amount 612; 426 Weighted average of last six semester values; 1. 3

4 As part of the analysis we examined the claim count and claim amount development triangles for each of the top ten private passenger automobile insurer groups in Nova Scotia. During the course of our review we identified one insurer with reported Bodily Injury claim counts and claim amounts over recent accident halfyears that appeared to be inconsistent with its reported claim counts and claim amounts over prior accident halfyears. We learned that the insurer (which we will refer to as Insurer A), changed the way it recorded (and reported to GISA) its Bodily Injury claims essentially not reporting claims for which it was believed that no loss (indemnity or ALAE) amounts would be paid. This change began during the first half of As respects Insurer A, without any adjustments to recognize its change in reporting, the claim count and claim amount development factors that we select, and hence the ultimate claim counts (frequency) and claim amounts (severity) that we select, would not be appropriate for the accident halfyears 2161and 2162 affected by the change. Claim Counts For accident years through 2152, we made no changes to the manner in which we selected development factors and ultimate claim counts as described earlier. For accident halfyears 2161 and 2162, we adjusted the Industry claim count triangle to remove Insurer A. For each semester, we then selected claim count development factors and ultimate claim counts for this semester based on the Industry data excluding Insurer A, added in the ultimate claim counts we selected for Insurer A 3, combined the two estimates of ultimate claim counts, and then backed into the claim count development factor for this semester. 2 We also found an insurer that changed the way it recorded and reported Bodily Injury claims to GISA beginning in 28. The effect was a reduction in reported claim frequency by approximately 75% and a significant increase in its reported claim severity. As this change in reporting has been consistent since 28, and our selected trend rate (as discussed below in the trend section) is based on data after 28, we made no adjustments for this insurer. 3 Insurer A s actuaries were unable to provide an estimate of the number of claims that would have been reported had no changes been made to claim reporting /recording practices. We, therefore, assumed Insurer A would have experienced the same change in claim frequency from 2152 to 2162 and from 2151 to 2161 as the rest of the Industry. 4

5 Claim Amounts For accident years through 2152, we made no changes to the manner in which we selected development factors and ultimate claim amounts as described earlier. For accident halfyears 2161 and 2162, we adjusted the Industry claim amount triangle to remove Insurer A. For each semester, we then selected claim amount development factors and ultimate claim amounts for this semester based on the Industry data excluding Insurer A, added in the ultimate claim amounts we selected for Insurer A 4, combined the two estimates of ultimate claim amounts, and then backed into the claim amount development factor for this semester. Exhibit 2, attached, presents our selected cumulative claim count and claim amount development factors which includes the noted adjustments for Insurer A. We note that as a result of these selected development factors and the actual experience that has emerged, our estimated ultimate claim counts and ultimate claim amounts have changed from our prior estimates, and these changes contribute to the changes in our selected trend rates. Selection of Trend Rates The identification of the underlying trend patterns over the experience period is challenging because factors such as statistical fluctuation in the data points, changes in the underlying exposure, or abnormal weather conditions, etc., can make the underlying trend patterns difficult to discern. For this reason, we model the data several different ways in an attempt to identify the underlying trends during the experience period over time periods that are longer than the experience period as a means of increasing the stability/reliability of the data being analyzed, but at the same time being responsive to changes in patterns that may have occurred, and with and without certain data points to improve our understand of the sensitivity of the calculated loss trend rate to the inclusion or exclusion of those points. 4 Insurer A s actuaries were unable to provide an estimate of the claim amounts that would have been reported had no changes been made to claim reporting /recording practices. We, therefore, assumed Insurer A would have experienced the same change in claim severity from 2152 to 2162 and 2151 to 2161 as the rest of the Industry. 5

6 Period Considered We present the experience by accident halfyear, spanning the period 221 to 2162, but in selecting past trend rates we give greater consideration to the experience over the more recent years. y Some coverages exhibit what is referred to as seasonality where claim costs (number of claims or claim amounts) incurred during the first half of a year are generally higher/lower than claim costs incurred during the second half of a year. In the coveragebycoverage discussion that follows, we state whether or not a seasonality parameter is applied. We note, however, that we find that seasonality may be significant for some, but not all time periods; or significant for loss cost, or severity, or frequency, but not for all three. Hence, for most coverages we present the measured trends with and without applying seasonality. Reforms The purpose of a reform parameter is to isolate and, in a sense, remove the impact that reforms had on the level of claim costs so that the underlying claim cost trend can be identified. For Bodily Injury, we give special consideration to the 281 to 211 period where challenges to the Minor Injury Regulations (MIR) may have had an impact on the claims experience, which was then followed by the changes to the MIR in April 21. In particular: o We give consideration to a possible change in reporting pattern that might have occurred beginning January 28 as a result of challenges to the Minor Injury Regulations in particular, the Decision by the Supreme Court of Nova Scotia to uphold the Minor Injury Regulation released on December 15, 29, and the Supreme Court of Canada s Decision on May 27, 21 to refuse leave to appeal the Decision. o We give consideration to Bill 52, an amendment to the Automobile Minor Injury Regulations of the Insurance Act, enacted on April 28, 21. We give consideration to the Fair Act Insurance Reforms enacted on April 1, 212, which introduced higher maximum benefit levels for Benefits subcoverages. 6

7 Effective April 1, 213, the DCPD coverage was introduced in Nova Scotia. We give consideration to this change in our selected trend rates for both Property Damage (which includes DCPD) and Collision. Data Points We give special consideration to data points that we consider to have a material impact on the measured trends. Of note is that for certain coverages, most notably for DCPD and Collision, there were large increases (spikes) in frequency in the 2151 accident period. We believe these increases to be at least in part (perhaps mostly) attributed to the relatively high amount of snow precipitation in the first quarter of 215. Consideration of Severity, Frequency, and Trend Patterns We consider the observed severity, frequency, and loss cost trend patterns. In so doing we consider the results of statistical tests that we apply. As respects the Adjusted Rsquare, we generally refer to values greater than 8% to be high, values between 4% and 8% to be moderate, and values below 4% to be low. We consider pvalues under 5% to be significant. The confidence interval range presented represents a 95% probability level range. Future Trend Rates In selecting future trend rates, if appropriate, we adjust our selected past trend rates after giving consideration to the reforms changes that have occurred over the recent past if there is evidence of new patterns emerging. A discussion of our selected trend rates follows. The various trend patterns that we review and associated statistical results are summarized in Exhibit 3 for each of frequency, severity, and loss cost. ****************************************************************** Our Selected Trend Rates Bodily Injury Based on our analysis as of June 3, 216, we selected a past and future loss cost trend rate of +2.5% and a level change parameter (for the reform) of 1.23 at April 21. 7

8 We estimate that during 2162 compared to the prior corresponding accident half year (2152) the frequency rate, the average severity, and the loss cost changed by approximately 7.3%, 9.1%, and 15.7%, respectively. This large frequency decline is consistent with the assumption that the high 2151 frequency rate is weather related. We estimate that the loss cost for the accident year ending December 31, 216 decreased by 4.2% over the loss cost for the accident year ending December 31, The following graphs display our estimate of the actual loss cost (average claim cost per vehicle), average severity (average claim cost per claim), and frequency rate (average claim incidence rate) over the period 221 through , 8 Claim Freq. Per Thousand 3 6, , , 3, 2, , Claim Frequency Per Thousand Period As depicted by the above graphs, subject to variability, severity sharply declined following the 23 reforms, but remained relatively flat from 25 through 291, after which, severity has been increasing. Frequency has exhibited a declining pattern over the last fifteen years, including a sharp decline in 28 and a lower rate of decline beginning 29. Loss cost sharply declined following the 23 reforms through to accident year 28, when it experienced an approximate 21% decline from 27 to 28. The 28 loss cost represents the low point over the last ten years. Loss cost has exhibited an increasing trend since 28. An increase in the minor injury cap (from $2,5 to $7,5) took effect on April 28, 21. In our study prepared for the Nova Scotia Superintendent of Insurance, Cost Implications of 5 These changes reflect the noted adjustments made for Insurer A. 8

9 Changes to the Minor Injury Regulations, dated May 12, 21, we estimated that Bodily Injury severity and loss cost would increase by approximately 17% as a result of the increase to the minor injury cap. We had since revised our estimate to +23% based on the experience that has emerged. Following a similar analysis as presented in our prior reports (and discussed more fully in the reform section), we continue to estimate the impact of the increase in the minor injury cap to be +23%. However, we note that the 212 severity is lower than the 292 severity, and so the reform impact is not evidenced in the 212 severity. We also note that indexing of the $7,5 minor injury cap is likely affecting the severity and loss cost trends. We further note that although we expected the April 1, 212 increase in the Benefits sub coverage limits to reduce the Bodily Injury claim costs (all else being equal), at this stage there is no clear evidence of this. As in our prior review, we do not adjust the data for this expected decrease in the Bodily Injury claim costs. The effect on Bodily Injury as a result of the changes to the Benefits sub coverage limit changes is discussed in our report prepared for the Board titled 211 Automobile Insurance Review Options Cost Impact, dated July 8, 211. In prior reports we stated that the cause of the sharp frequency decline in 28 is not clear. Given the unexplained sharp decline, we continue to consider 28 to be a low point and note that it coincides with a change in the Bodily Injury loss cost trend pattern. The measured severity, frequency, and loss cost trends, associated Adjusted Rsquare values, p values, and confidence intervals over various trend measurement periods ending 2162 and 2161 (in consideration of the relatively higher degree of uncertainty surrounding the estimated loss cost for 2162), including a +23% adjustment to reflect our estimate of the impact of the April 21 reforms, with and without seasonality, and with and without the 212 data point which we find to be low 6 for severity, are presented in Exhibit Although the minor injury reforms with a higher cap amount were introduced effective April 1, 21, the 212 severity estimate is less than the 291 severity estimate. For this reason we consider the measured severity trend rates with and without this data point. 7 For additional information, we also present in Exhibit 3 the trend rates excluding the 281 data point as it appears to be low and the 2151 data point as it appears to be high. 9

10 We make the following observations about these measured trends. The severity trends, beginning 291 and onward and ending 2161 and 2162, with seasonality (which we find to be significant), and including 212, generally fall in the range of approximately +5.% to +6.5% with moderate Adjusted Rsquares and significant pvalues, with trends ending 2161 higher than those ending When the 212 data point is excluded, the measured trends beginning 291 through 211 are modestly lower, with slightly higher Adjusted Rsquares and lower (more significant) pvalues. We continue to select a severity trend rate of +5.%, based on the measured trends beginning 291. The measured frequency trends ending 2162, with seasonality (which we find to be significant), generally fall in the range of 1.% to 3.5% with moderate Adjusted Rsquares and generally significant pvalues. The more recent of these measured trends (i.e., beginning 211 through 2112) with significant pvalues fall within the range of 1.8% to 2.5%. We select a trend rate of 2.%. Based on the separate frequency and severity selections we select a past loss cost trend rate of +3.% (rounded); one half percentage point higher than our prior selection. As we see no clear evidence of a recent change in trend pattern, we select a future loss cost trend of +3.%. Property Damage (and DCPD) DCPD was introduced on April 1, 213. The data we analyze includes both Property Damage and DCPD experience. Based on our analysis as of June 3, 216, we selected a past loss cost trend of +3.% through to April 1, 213 and a loss cost trend rate of +7.% for the period beginning April 1, 213; with no reform parameter. We estimate that during 2162 compared to the prior corresponding accident half year (2152) the frequency rate, the average severity, and the loss cost changed by approximately +7.6%, +1.5%, and +9.2%, respectively. We estimate that the loss cost for the accident year ending 1

11 December 31, 216 decreased by.8% over the loss cost for the accident year ending December 31, 215. The following graphs display our estimate of the actual loss cost (average claim cost per vehicle), average severity (average claim cost per claim), and frequency rate (average claim incidence rate) over the period 221 through , 35 Claim Freq. Per Thousand 14 5, , 3, 2, Claim Frequency Per Thousand , Period The historical data points (as depicted in the above graphs) show that, subject to variability: (a) following a period of relative flatness (23 to 211) severity has exhibited an upward trend, (b) following a period of a relatively flat trend between 27 and the first half of 213, frequency increased in the second half of 213, and has since been rising, including a sharp spike in the first half of 215, which may be related to the noted weather conditions, and (c) loss cost has exhibited an upward trend, including a relatively large increase in the second half of 213 and a spike in the first half of 215 that may be weather related. We assume that the introduction of DCPD in April 213 caused a shift in claims from Collision to DCPD and this explains the increase in frequency and loss cost in 213. The measured severity, frequency, and loss cost trends, associated Adjusted Rsquare values, p values, and confidence intervals over various trend measurement periods beginning with and without seasonality, and with and without the 2151 data point are presented in Exhibit 3. 11

12 As in our prior review, we measure the trend rates before and after DCPD was introduced (without a reform parameter) as there appears to be a change in the loss cost trend rate following the introduction of DCPD. For the period before DCPD was introduced we consider the loss cost trends as they have higher Adjusted Rsquare values than the separate severity and frequency trends. The measured loss cost trends, beginning 221 through to 261 ending 2122, with seasonality, which we find to be significant, cluster around +3%, with high Adjusted Rsquares and significant pvalues for time. We select a loss cost trend through of +3.%, the same as our prior selection, to apply to the Property Damage loss experience both before and after the introduction of DCPD. We consider the trend rates after DCPD was introduced based on combined 8 Property Damage and DCPD experience. While the post reform data is limited (particularly with the exclusion of the 2151 data point), we see statistical evidence (high Adjusted Rsquares and significant pvalues) of an increase in the trend pattern for loss cost after the introduction of DCPD as measured over the more recent trend periods from 2132 through to The measured loss cost trend over the period 2132 to 2162, excluding 2151, without seasonality, is + 9.1% with a high Adjusted R square and significant pvalue for time. We select a loss cost trend rate of +9.% for the period beginning April 1, 213; an increase of two percentage points from our prior selection. We select a past loss cost trend of +3.% through to April 1, 213 and a loss cost trend rate of +9.% for the period beginning April 1, 213; with no reform parameter. Benefits Disability Income Based on our analysis as of June 3, 216 we selected a past and future loss cost trend rate of +.% with a level change parameter (for the reform) at April 212 of We estimate that during 2162 compared to the prior corresponding accident half year (2152) the frequency rate, the average severity, and the loss cost changed by approximately 1.%, 8 DCPD loss cost is approximately 95% of the total Property Damage and DCPD loss cost. 12

13 21.9%, and 22.7%, respectively. We estimate that the loss cost for the accident year ending December 31, 216 decreased by 9.3% over the loss cost for the accident year ending December 31, 215. The following graphs display our estimate of the actual loss cost (average claim cost per vehicle), average severity (average claim cost per claim), and frequency rate (average claim incidence rate) over the period 221 through , 3 Claim Freq. Per Thousand 25 25, , 15, 1, Claim Frequency Per Thousand , Period The historical data points indicate a considerable amount of variability, particularly for severity and loss cost (due to severity). Severity has generally exhibited an increasing pattern, including a decline over the 27 to 29 period, a sharp increase in 21 (+9% from 291 to 211), and a sharp increase in the second half of 212 (immediately following the reform). Frequency has exhibited a downward trend that leveled off somewhat (a less steep decline) beginning in 21. Loss cost, following a steep decline in 28 and 29, has generally been increasing since; and, like severity, has experienced sharp increases in 21 and the second half of 212. Effective April 1, 212, the Disability Income weekly benefit was increased from $14 to $25, and for unpaid housekeepers from $7 to $1. Our estimates of the reform parameter are wide ranging with generally weak pvalues over most the time periods beginning 221, which we believe is a reflection of the high degree of variability in the Disability Income severities. Given this, we continue to select the reform parameter of 1.43 that we calculated and presented to the Board in our July 211 report, as we selected in our prior review. 13

14 The measured severity, frequency, and loss cost trends, associated Adjusted Rsquare values, p values, and confidence intervals over various trend measurement periods, with and without seasonality, with preapril 212 severities increased by 43% are presented in Exhibit 3. The measured severity trends beginning 221 through 292 ending 2162 without seasonality (which is not significant) generally range from approximately +5% to +7%, with moderate to low Adjusted Rsquares, but significant pvalues. Given the volatility in severity (and wide confidence intervals) since the April 21 reforms were introduced, we continue to select the severity trend based on measured trends over the longer term periods. We select a severity trend rate of +5.5%; the same as our prior selection. The measured frequency trends with seasonality (which is significant) generally range from approximately 3% to 5.5%, with generally moderate to high Adjusted Rsquares and significant pvalues. The trend rates over the longer term periods (ten to fifteen years) cluster around 5.5% with high Adjusted Rsquares. Like for severity, we select a frequency trend based on measured trends over the longer time periods. We select a frequency trend of 5.5%. We also note that all of the measured loss cost trends have low (near zero) Adjusted Rsquares and pvalues that are not significant for time. As a result, based on our selected severity trend rate of +5.5% (and level change parameter of 1.43 at April 212) and selected frequency trend of 5.5%, we continue to select a past loss cost trend rate of +.% (rounded), with a level change parameter at April 212 of 1.43, the same as our prior review. As we see no clear statistical evidence of a recent change in trend pattern, we select a future loss cost trend of +.%. Benefits Medical/Rehab Based on data as of June 3, 216, we selected a past and future loss cost trend rate of +3.% with level change parameters (for reforms) at April 21 of 1.25 and April 212 of We estimate that during 2162 compared to the prior corresponding accident half year (2152) the frequency rate, the average severity, and the loss cost changed by approximately +7.2%, 14

15 12.6%, and 6.3%, respectively. We estimate that the loss cost for the accident year ending December 31, 216 increased by 1.6% over the loss cost for the accident year ending December 31, 215. The following graphs display our estimate of the actual loss cost (average claim cost per vehicle), average severity (average claim cost per claim), and frequency rate (average claim incidence rate) over the period 221 through , 1 Claim Freq. Per Thousand 45 7, , , 4, 3, , , Claim Frequency Per Thousand Period The historical data points indicate a considerable amount of variability, particularly for severity and loss cost (due to severity). Severity has generally exhibited an upward trend, but with relatively sharp increases in 21 and in 212 and, following declines in 213 and 214, increases in 215 (modest) and 216 (larger). Frequency has trended upward since 29. Loss cost has generally been increasing since 28; and, like severity, experienced relatively sharp increases in 21 and 212, and following declines in 213 and 214, increases in 215 and 216. The measured severity, frequency, and loss cost trends, associated Adjusted Rsquare values, p values, and confidence intervals over various trend measurement periods, with and without seasonality, and with level changes for the reforms are presented in Exhibit 3. We attribute the severity and loss cost increases in 21 to the April 21 reforms. We select a level change parameter of 1.24 at April 21 based on the time period 281 to 2162 which has the highest Adjusted Rsquare. 15

16 Due to the volatility of the claim experience and the short time gap between the April 21 and April 212 reforms, the degree to which the April 212 reforms contributed to severity and loss cost increases in 212 is not clear. However, we continue to assume that the April 212 reforms led to a 15% increase in claim costs 9 and note that the majority of the experience that will be considered in upcoming rate applications will be after the April 212 reforms. The severity trends, with seasonality, and with losses prior to April 212 and losses prior to April 21 adjusted upward by 15% and 43% (1.43 = 1.24*1.15), respectively, to reflect the two reforms, are close to %, with low Adjusted Rsquares and pvalues that are not significant for time. We select a severity trend of +%. The frequency trends beginning 291 through 2131 ending 2162 range from approximately +3.5% to +5.5%. The more recent measured frequency trends are at the higher end of this range, with moderate Adjusted Rsquares and pvalues that are significant for time. We select a frequency trend of +5%. We, therefore, select a loss cost trend of +5.%, with an April 21 reform adjustments of +24% and an April 212 reform adjustment of +15%. The selected trend is two percentage points higher than our prior selection. Benefits Funeral Based on data as of June 3, 216, we selected a past and future loss cost trend rate of 5.% with a level change (for reform) parameter at April 212 of 2.5. We estimate that during 2162 compared to the prior corresponding accident half year (2152) the frequency rate, the average severity, and the loss cost changed by approximately +4.4%, +11.8%, and +16.7%, respectively. We estimate that the loss cost for the accident year ending December 31, 216 increased by 5.6% over the loss cost for the accident year ending December 31, 215. Effective April 1, 212, the Funeral maximum benefit was increased from $1, to $2,5. 9 This is the same adjustment for the April 212 reforms as in our prior review, based on our study for the Board in our report prepared July

17 The following graphs display our estimate of the actual loss cost (average claim cost per vehicle), average severity (average claim cost per claim), and frequency rate (average claim incidence rate) over the period 221 through ,5 Claim Freq. Per Thousand 4, 3,5 3, 2,5 2, 1,5 Claim Frequency Per Thousand 1, Period The historical data points indicate a considerable amount of variability. Severity has generally exhibited an upward trend since 25 with a sharp increase coincident with the reform. Frequency has generally trended downward. The measured loss cost trends, associated Adjusted Rsquare values, pvalues, and confidence intervals over various trend measurement periods, with and without seasonality, and with a factor of for the April 212 reform applied to the data are presented in Exhibit 3. The measured loss cost trends without seasonality (which is not significant) beginning 221 through 271 ending 2162, with a 2121 level change factor of 2.5 that have low to moderate Adjusted Rsquares and significant pvalues for time, are in the approximate range of 9% to 13%. Given the low claim volume, we select the measured loss cost trend over the time period 221 to 2162 at 9%. 1 As presented in Exhibit 3, the measured reform factors with significant pvalues are in the range of the 2.5 factor we apply to the data. 17

18 We select a past and future loss cost trend of 9.%, a decrease from our prior selection of 5%. Benefits Death Based on data as of June 3, 216, we selected a past and future loss cost trend rate of.% with a level change (for reform) parameter at April 212 of 2.5. We estimate that during 2162 compared to the prior corresponding accident half year (2152) the frequency rate, the average severity, and the loss cost changed by approximately +1.8%, +51.4%, and +67.7%, respectively. We estimate that the loss cost for the accident year ending December 31, 216 increased by 48.5% over the loss cost for the accident year ending December 31, 215. Effective April 1, 212, the Death maximum benefit was increased from $1, to $25, for spouses and headofhousehold, and from $2, to $5, for dependents. The following graphs display our estimate of the actual loss cost (average claim cost per vehicle), average severity (average claim cost per claim), and frequency rate (average claim incidence rate) over the period 221 through , Claim Freq. Per Thousand 2 2 3, , 2, 15, 1, Claim Frequency Per Thousand 5, Period The historical data points indicate a considerable amount of variability. Severity has generally exhibited an upward trend with a sharp increase coincident with the reform. Frequency has generally trended downward. 18

19 Due to the volume and variability of the claim experience, we consider only the observed loss cost trends. The measured loss cost trends, associated Adjusted Rsquare values, pvalues, and confidence intervals over various trend measurement periods, with no seasonality, and with a factor of for the April 212 reform applied to the data are presented in Exhibit 3. The measured loss cost trends beginning 221 through to 271 ending 2162 without seasonality (which is generally not significant) have low to moderate Adjusted Rsquares and range from 5.5% to 12%; with the larger negative trends coming from the more recent time periods. Given the low claim volume, we select the measured loss cost trend over the time period 221 to 2162 at 5.5%. We select a past and future loss cost trend of 5.5%; a decrease from our prior selection of.%. Benefits Total Based on our review of the trends for the Benefits subcoverages described above, we select an approximate past and future loss cost trend of +3.%(rounded) for this coverage. Collision Based on data as of June 3, 216, we selected a past loss cost trend rate of +.% through to April 1, 213, and a loss cost trend rate of +2.5% for the period beginning April 1, 213. We estimate that during 2162 compared to the prior corresponding accident half year (2152) the frequency rate, the average severity, and the loss cost changed by approximately +7.4%, +2.4%, and +9.9%, respectively. We estimate that the loss cost for the accident year ending December 31, 216 decreased by 2.6% over the loss cost for the accident year ending December 31, 215. As we noted earlier, DCPD was introduced April 1, 213 and as we discuss more fully below, the introduction of DCPD appears to have affected the Collision claim experience. 11 As presented in Exhibit 3, as the measured reform factors are wide ranging, we continue to apply a factor of 2.5 for the April 212 reforms. 19

20 The following graphs display our estimate of the actual loss cost (average claim cost per vehicle), average severity (average claim cost per claim), and frequency rate (average claim incidence rate) over the period 221 through , 5 Claim Freq. Per Thousand 45 6, , 4, 3, 2, , Claim Frequency Per Thousand Period As noted in our prior report, the sharp increase in the 2151 frequency is likely attributed to a high snowfall level, and as a result, the decline in the 2161 frequency is attributed to a return to more average snowfall levels. The historical data points show that, subject to variability, severity has generally exhibited an upward trend including double digit increases in 2131, 2132, 2141 and 2161 (over the same half of the prior year). Frequency has generally declined since the 2729 period, save for a 42% increase in the first half of 215, which we assume is attributed to the noted weather conditions. Loss cost exhibited a period of relatively flat trend beginning around 28, but an upward trend over the more recent time periods, including a spike in As in our prior review, we measure the trend rates before and after DCPD was introduced (without a reform parameter) as we see evidence of a change in the trend pattern with the introduction of DCPD. The measured severity, frequency, and loss cost trends, associated Adjusted Rsquare values, p values, and confidence intervals over various trend measurement periods, with and without seasonality, and with and without the 2151 data point are presented in Exhibit 3. 2

21 The measured trends beginning 221 through 292 ending 2122, before DCPD was introduced, with significant pvalues for the time and moderate Adjusted Rsquares, cluster close to +2.5%. The measured frequency trends beginning 271 (where we find there to be a change in pattern) through 292 ending 2122, before DCPD was introduced, with significant pvalues for the time and moderate Adjusted Rsquares generally range from 2.5% to 3.5%. We select a loss cost trend rate of.5% (severity: +2.5%; frequency 3.%) through to April 1, 213, before DCPD was introduced. As the data after DCPD was introduced is limited (particularly with the exclusion of the 2151 data point), with poor fits for frequency, we consider the loss cost trends. We see statistical evidence (moderate Adjusted Rsquares and significant pvalues) of an increase in the trend pattern for the loss cost as measured over the more recent trend periods since DCPD was introduced. The measured loss cost trend from 2132 to 2162, without 2151, with a significant pvalue for the time, is approximately +4.5%. We select a loss cost trend of +4.5% for the time period after DCPD was introduced. We, therefore, select a past loss cost trend of.5% through to April 1, 213 and a loss cost trend rate of +4.5% for the period beginning April 1, 213; a onehalf point lower trend through April 1, 213 and a two point higher trend beginning April 1, 213 than our prior selections. Comprehensive Based on data as of June 3, 216, we selected a past and future loss cost trend rate of +6.%. We estimate that during 2162 compared to the prior corresponding accident half year (2152) the frequency rate, the average severity, and the loss cost changed by approximately +1.3%, +17.%, and +18.6%, respectively. We estimate that the loss cost for the accident year ending December 31, 216 increased by 9.5% over the loss cost for the accident year ending December 31, 215. The following graphs display our estimate of the actual loss cost (average claim cost per vehicle), average severity (average claim cost per claim), and frequency rate (average claim incidence rate) over the period 221 through

22 14 2,5 8 Claim Freq. Per Thousand , ,5 1, Claim Frequency Per Thousand Period The historical data points (as depicted in the above graphs) show that severity has generally exhibited an upward trend, although somewhat of a flat trend between 23 and 21, and, following about rather sharp decline in 211, a higher trend thereafter. Frequency has been more variable, but generally increasing since 25; like Property Damage and Collision, the Comprehensive frequency rose in the first half of 215 (by 17%). Subject to variability, the loss cost has exhibited an upward trend since 24. The measured severity, frequency, and loss cost trends, associated Adjusted Rsquare values, p values, and confidence intervals over various trend measurement periods, with and without seasonality, with and without the 2151 data point, are presented in Exhibit 3. The measured severity trends with seasonality (which is significant) data point over the periods beginning 2111 through 2131 ending 2162 cluster close to +7.% with strong regression statistics. We select a severity trend of +7.%. Although, as noted above frequency has exhibited a positive trend since 25, it has done so with considerable variability. Over the same trend period upon which our selected severity trend is based, without seasonality (which is not significant), the measured frequency trends regression statistics are weak. We, therefore, select a frequency trend of +.%. As a result, we select a past loss cost trend of +7.%; a one percentage point increase over our prior selection. 22

23 As we see no clear statistical evidence of a recent change in trend pattern, we select a future loss cost trend of +7.%. Specified Perils For reasons of data volume and the nature of the coverage, we select the same past and future loss cost trend rate as we do for Comprehensive, +7.% for the past and the future. All Perils Based on data as of June 3, 216, we selected a past and future loss cost trend rate of +.5%. We estimate that during 2162 compared to the prior corresponding accident half year (2152) the frequency rate, the average severity, and the loss cost changed by approximately +14.1%, +11.6%, and +27.3%, respectively. We estimate that the loss cost for the accident year ending December 31, 216 increased by 7.3% over the loss cost for the accident year ending December 31, 215. The following graphs display our estimate of the actual loss cost (average claim cost per vehicle), average severity (average claim cost per claim), and frequency rate (average claim incidence rate) over the period 221 through , 9 Claim Freq. Per Thousand 3 4,5 8 4, ,5 3, 2,5 2, 1, , Claim Frequency Per Thousand Period 23

24 The historical data points (as depicted in the above graphs) show considerable variability, with severity generally increasing more sharply following a decline in the period and frequency exhibiting a flatter trend pattern but declining beginning in 211; like Property Damage & DCPD, Collision, and Comprehensive, frequency rose in the first half of 215 (by 22%). The measured severity, frequency, and loss cost trends, associated Adjusted Rsquare values, p values, and confidence intervals over various trend measurement periods, with and without seasonality, and with and without the 2151 data point, are presented in Exhibit 3. The measured severity trends without seasonality (which is not significant), increase rather significantly beginning in 211 with high Adjusted Rsquare values and significant pvalues. We select a severity trend of +7.5% based on the measured trends beginning 2111 through 2131 ending 2162 with the high Adjusted Rsquares that range from approximately +7% to +8%. The measured frequency trends without seasonality (which is not significant), without the 2151 data point (for which the regression statistics are stronger) more sharply decline with trend periods beginning 212 through 2112, ending 2162 with moderate Adjusted Rsquare values and significant pvalues. We select a frequency trend of 4.% based on the measured trends beginning 2111 and As a result, we select a past loss cost trend of +3.5% (rounded); a 3. percentage point increase over our prior selection. As we see no clear evidence of a recent change in trend pattern, we select a future loss cost trend of +3.5%. Underinsured Motorist For reasons of data volume and the nature of the coverage, we select as the past loss cost trend rate, the severity trend rate of +5% that approximately underlies our selected Bodily Injury severity trend rate. 24

25 Uninsured Auto Based on data as of June 3, 216, we selected a past and future loss cost trend rate of +8.%. We estimate that during 2162 compared to the prior corresponding accident half year (2152) the frequency rate, the average severity, and the loss cost changed by approximately 5.9%, 9.2%, and 14.5%, respectively. We estimate that the loss cost for the accident year ending December 31, 216 decreased by 2.9% over the loss cost for the accident year ending December 31, 215. While these increases and decreases could be viewed as an indication of the inherent variability in the Uninsured Auto claim experience (due to the low claim volume), it must also be noted that the degree of uncertainty surrounding the recent accident half year estimates is quite high. The following graphs display our estimate of the actual loss cost (average claim cost per vehicle), average severity (average claim cost per claim), and frequency rate (average claim incidence rate) over the period 221 through , 1 Claim Freq. Per Thousand 14 35, , , 2, 15, Claim Frequency Per Thousand 4 1, 2 5, Period The historical data points show considerable variability, with severity generally increasing particularly beginning in 25 and frequency generally declining through 29 and increasing thereafter. Correspondingly, loss cost has been generally increasing since 29; with a very large spike in Due to the volume and variability of the claim experience, we consider only the observed loss cost trends. 25

26 The measured loss cost trends, associated Adjusted Rsquare values, pvalues, and confidence intervals over various trend measurement periods, with seasonality for loss cost, with and without the very high 2132 data point are presented in Exhibit 3. We find seasonality to be generally significant for loss cost. We select a past and future loss cost trend rate of +9.% based on the measured trends beginning 26 to 27 ending 2162, excluding the 2132 data point, due to the combination of the higher Adjusted Rsquares and narrower confidence interval than shorter more recent time frames. 26

27 Selected Trend Rates Summary The following table presents our selected past and future annual loss cost trend rates as of December 216. Coverage Past Future Bodily Injury +3.% +3.% Property Damage & DCPD +3.% +9.%* AB Disability Income +.% +.% AB Medical/Rehab +5.% +5.% AB Funeral 9.% 9.% AB Death 5.% 5.% ABTotal +3. % +3.% Collision.5% +4.5%* Comprehensive +7.% +7.% Specified Perils +7.% +7.% All Perils +3.5% +3.5% Underinsured Motorist +5.% +5.% Uninsured Auto +9.% +9.% *As of April 1, 213 For comparison purposes, the following table presents our prior selected past and future annual loss cost trend rates as of June 216 presented in our prior report. Coverage Past Future Bodily Injury +2.5% +2.5% Property Damage & DCPD +3.% +7.%* AB Disability Income +.% +.% AB Medical/Rehab +3.% +3.% AB Funeral 5.% 5.% AB Death +.% +.% ABTotal +2.% +2.% Collision +.% +2.5%* Comprehensive +6.% +6.% Specified Perils +6.% +6.% All Perils +.5% +.5% Underinsured Motorist +5.% +5.% Uninsured Auto +8.% +8.% *As of April 1,

28 Reforms In accordance with Bill 52, Bodily Injury claims that occur on or after April 28, 21 are subject to a minor injury cap of $7,5. Bill 52 changed both the amount of the cap and the definition of a minor injury. Following a Hearing on the matter, the Board accepted an initial reform adjustment factor of 1.17 for Bodily Injury, and ordered that the data be monitored as it emerges so as to measure the change, if any, in the loss trend rate and the actual change in loss costs due to Bill 52. We have reviewed the Industry Bodily Injury experience that has emerged since the Bill 52 reforms were introduced to determine if the initial reform factor of 1.17 should be amended. The Bodily Injury frequency experience over the last fifteen years is presented in the graph below. As depicted in the graph, we do not observe there to have been a change in the frequency level as a result of the introduction of Bill 52 in the first half of 21. The frequency rate shows a pattern of decline, with a steep decline in 28. And although the frequency trend rate has been somewhat flatter (smaller negative trend rate) since 29, just before the reforms were introduced, it is not clear that this smaller negative trend rate is a result of the reforms or a leveling off of the decline in frequency that has been observed in other provinces. 28

29 9 Claim Freq. Per Thousand Claim Frequency Per Thousand The Bodily Injury severity experience over the last fifteen years is presented in the graph below. As depicted in the graph, we observe there to have been an increase in the severity level as a result of the introduction of Bill 52 in the first half of 21. So as to avoid any distortion from the Bill 1 (232) reforms, we calculate a reform factor for this change in the severity level based on the experience period beginning 241. Including a parameter for seasonality, which we find to be significant, we select a parameter of 1.23 based on the time period 251 to excluding the 212 data point and including seasonality, with the highest Adjusted Rsquare (92%), the same as our prior selection. 29

30 7, 6, 5, 4, 3, 2, 1, As noted earlier, the $7,5 minor injury cap is indexed. The cap increased to $7,596 on January 1, 212; to $8,1 on January 1, 213; to $8,213 on January 1, 214 $8,352 on January 1, 215, $8,385 on January 1, 216 and $8,486 on January 1, 217. In the case of MedicalRehabilitation, based on our review of the experience to date, and integration of reform parameters within our loss trend models, we continue to find that there to have been an increase in MedicalRehab severity following the April 21 Bodily Injury reforms. We select a level change parameter of 1.24 at April 21 based on the time period 281 to 2162 which has the highest Adjusted Rsquare at 87% and significant pvalues. (The associated loss cost trend that we select is +5.5%.) Our selected April 21 reform factor of 1.24 is slightly lower than our prior selection of

31 In addition to the possible impact of Bill 52 on the MedRehab costs, the Fair Insurance Reforms introduce higher Benefit limits effective April 1, 212 as presented in the following table: Benefit Category Previous Benefit New Benefit (as of April 1, 212) Medical and Rehabilitation Expenses $25, $5, Funeral Expenses $1, $2,5 Death Benefits Head of Household $1, $25, Spouse of Head of Household $1, $25, Dependent $2, $5, Loss of Income $14/week $25/week Principal Unpaid Housekeeper $7/week $1/week However, we do not find the data to show any statistically significant measures for the change in benefit level for the April 212 reforms at this time for MedicalRehabilitation. This may be due to the inherent volatility of the data or because these two changes (April 21 and April 212) are so close together. Our April 212 reform estimate remains at 1.15 as per our original estimate prepared in July 211 for the Board. In the case of Disability Income, as presented in Exhibit 3, while we measure a significant severity level change factor of 1.52 (over the latest fifteen year period and excluding the 211 data point) for the April 212 reforms, as the level change value varies depending upon the time frame considered, we choose to continue to select our original April 212 reform factor of 1.43 and apply this directly to the data. In Exhibit 3 we present the reform factors that we calculate over various time periods for Funeral and Death Benefits. Given the limited data, and the range of factors that we calculated, our selected reform factors for Funeral and Death Benefits remain unchanged at 2.5 and 2.5, respectively, at this time. 31

32 In our report prepared for the Board dated July 211, we estimated an approximate Benefits loss cost reform factor of 1.3 and this increase to the Benefits loss cost of approximately $11 would be partially offset by a reduction to the Bodily Injury loss cost of $7, for an overall increase of approximately $4. As these changes were only introduced midway through the first half of 212, the actual cost impact of these reform remains uncertain. Our reform impact estimates for these benefit level reforms remains unchanged. As discussed in this report, there appears to have been a change in Collision beginning in 2131 due to the introduction of the DCPD coverage in April 213. We find there to be an increase in the trend rate for both these coverages coincident with the introduction of DCPD. Exhibits In the Exhibit 1 we present the historical loss cost, severity and frequency data points by accident half year over the fifteen year period 221 to 2162, as well as the data points for each coverage. In Exhibit 2 we present our selected cumulative claim count and claim amount development factors. In Exhibit 3 we present the summary of the loss trend rates we have calculated over various time periods along with the associated regression statistics. Paula Elliott, FCAS, FCIA Theodore J. Zubulake, FCAS, FCIA 32