| The aging of the population in Hong Kong is getting serious.The need to improve the mandatory provident fund system,which is currently the main retirement protection mechanism,becomes pressing.This paper hopes to provide future individuals or government departments who make an attempt to improve the Hong Kong Mandatory Provident Fund system with a workable assessment mechanism and valid research data.This will help improve the quality of life in the community and promote social stability in Hong Kong.First,this paper used the three-factor model,the Sharpe ratio and the Treynor ratio to examine the efficacy of MPF returns in Hong Kong.No significant deficiency in sectors other than money market funds was found.Second,this paper used two traditional models,the Treynor and Mazuy Model and the Henriksson and Merton Model,to measure the timing ability of MPF managers with daily return data.Daily return data were used because high sampling frequency promotes resolution which in turn benefits the accuracy of the assessment model.According to the traditional models,about 69% of MPF fund managers did not demonstrate timing actions.For most of the rest of the managers,their timing actions had not only failed to promote the interests of the funds' trustees,but even had done harm to their interests.Therefore,this paper tried to find out,as a follow-up study,whether the unsatisfactory performance of the fund managers was really the case or whether the assessment mechanism needs to be tailored according to the specific needs of MPF.Since the assessment mechanism on fund manager's timing ability this paper seeks is for screening able managers for mandatory provident funds,it naturally follows that“the target of the screening mechanism must be someone who is long-term-return oriented” is a premise of the search.After a close examination,it was discovered that,with high return data sampling frequency,even managers with the best timing abilities will not be rated by the traditional models as “demonstrating timing ability with statistical confidence”.After an in-depth exploration,it was discovered that this was caused by data points called “invaders” hereafter on the excess investment portfolio return vs.excess market portfolio return Cartesian coordinate plane.Invaders are data points with the return for the sampling period opposite to the long-term market trend.They are daily,monthly,or annual data points in long-term upswings with negative excess market returns,or in long-term downswings with positive excess market returns.Aboriginals are data points with the return for the sampling period in line with the long-term market trend.They are daily,monthly,or annual data points in long-term upswings with positive excess market returns,or in long-term downswings with negative excess market returns.In accordance with the concept of the two traditional models,a fund manager's timing ability is reflected in predicting correctly the market conditions in each of the sampling period concerned and configuring the betas according to the prediction.High beta for a predicted positive market return in the coming sampling period and low beta for a predicted negative market return.Only a sampling period with positive excess market return should have its beta configured to a high value.Therefore,in the excess market portfolio return vs.excess investment portfolio return Cartesian plane,only high-beta data points should be found on the right side of the excess investment portfolio return axis.However,long-term-return oriented managers with the best timing abilities will set the beta of the fund to high value for each sampling period within a long-term bullish market,no matter the sampling period ends up with positive market return,which is in line with the long-term market condition and is thus an aboriginal;or the sampling period ends up with negative market return,which is opposite to the long-term market condition and is thus an invader.They will also set the beta of the fund to low value for each sampling period within a long-term bearish market,no matter the sampling period ends up with negative market return,which is in line with the long-term market condition and is thus an aboriginal;or the sampling period ends up with positive market return,which is opposite to the long-term market condition and is thus an invader.Therefore,aboriginals will show up with high beta values on the right side and with low beta values on the left side of the excess investment portfolio return axis respectively.Invaders will show up with low beta values on the right side and with high beta values on the left side of the excess investment portfolio return axis respectively.Invaders with high beta values on the left side join aboriginals with high beta values on the right side to form a straight line with high slope.Aboriginals with low beta values on the left side join invaders with low beta values on the right side to form a straight line with low slope.In this way the data points of a long-term-return oriented manager with the best timing abilities shape a pair of crossing straight lines,and thus fail to meet the expectation of the traditional models for managers with the best timing abilities either as a parabola or a polyline facing top left.What's more,increasing the sampling frequency of return data will increase the percentage of invaders in the data set,thus making the data graph of fund managers with good timing ability deviate more from the expectations of the two traditional models.The root cause of the problem rests in the fact that the traditional models use a sampling-period-return oriented strategy to set the taregt of screening.When the sampling period changes from each year as in the past to become each day as nowadays,a sampling-period-return oriented strategy becomes a short-term-return oriented one in effect.Thus if the sampling frequency of returns is raised without matching adjustments made in other aspects,the practice of seeking for mid-to-long-term-return oriented managers with the best timing abilities in the past will become the seeking for shortterm-return oriented managers with the best timing abilities nowadays.To hold on to a long-term-return orientation after sampling frequency is raised,an explicit switch of the assessment mechanism at the base level from a sampling-period-return oriented strategy to a long-term-return oriented one is a must.What is proposed in this paper is to have the determinant of target beta level,a corner-stone of the assessment mechanism,be changed from the sign of the return of the sampling period to the longterm market trend.Since the invaders have a serious impact on the effectiveness of the two traditional models at high sampling frequencies,this paper makes an attempt to improve the capability of the traditional models by removing the invaders from the dataset before performing regression analysis on the dataset.The assessment results of the fund managers' timing abilities from these fine-tuned versions of the two traditional models thus obtained have shown some improvements over the original versions.In order to achieve further improvements,this paper followed the thread of discussion above to design a new timing ability assessment model called the Bull and Bear Model hereafter.A fundamental difference between the new model and the traditional models is that the former uses the judgment of the long-term market conditions as the determiner of the beta value level for that sampling period while the latter uses the prediction on the excess market return of the sampling period.According to the criteria set in the traditional models,under the sampling frequency of daily return data,if the chance of a downward adjustment in tomorrow's market condition is more than half,no matter whether it is in a long-term bullish market or not,at the close of the market a fund manager with timing ability should change the beta of the portfolio to low.According to the criteria set in the new model,under the judgment of a long-term bull market,even if the chance of a downward adjustment in the market conditions of the following trading day is more than half,the portfolio's beta should be kept high.The idea underlying the new model is as follows: The beta value reflects the intrinsic characteristics of the assets held by the fund portfolio.Theoretically,within the period that beta value is maintained,the more samples taken the higher the statistical reliability of the estimate of the beta.By raising the sampling frequency more samples can be taken in the same period thus increasing the reliability of the results.However,in the two traditional models,the fund manager's timing ability is reflected in predicting correctly market conditions of the sampling period and configuring beta accordingly.As a result,the frequency of beta target value reviewing is tied to the sampling frequency of returns.Raising the sampling frequency of returns will increase the frequency of beta target value reviewing.The increase in the frequency of beta target value reviewing will make the target of screening sought by the traditional models shift from long-term-return oriented managers to short-term-return oriented ones.A sampling-period-return oriented approach becomes a short-term-return oriented approach in effect.Even for the same fund manager,correctly predicting for each trading day in the coming year whether the market return will go up or down is much much more difficult than predicting whether the annual market return in the coming year will go up or down.Thus the speculative nature of beta-reconfiguration in the former is obvious.Therefore,the increase in beta reviewing frequency will bias the two traditional models to favor speculators.In order to keep Treynor's intention of keeping beta being reviewed not more frequently than once a year(Treynor 1966),and to be in line with the purpose of screening for MPF managers,there is a need to decouple the beta reviewing frequency from the return sampling frequency in the assessment mechanism.The Bull and Bear Model uses long-term market conditions to determine beta configuration,so it is long-term-return oriented rather than sampling-period-return oriented.Therefore,increasing the frequency of return sampling from annual to daily does not require an increase in the beta value reviewing frequency.In fact successive beta value reconfiguration may be separated by several years since a bull market may last for several years.At high sampling rates invaders occurred in the dataset,which is beyond the expectations of the traditional models.Regression calculations involving those lowbeta aboriginals formed in the bear markets and thus are on the left side of the excess investment portfolio return axis and are subject to interference from invaders formed in the bull markets extending to the left following a high beta trend from the right.Similar interferences from invaders formed in the bear market happen to those aboriginals formed in the bull market.Since the Bull and Bear model groups data points according to moments of longterm market reversal as group boundaries in the time dimension,all data points in the same regression dataset belong to the same bull market or bear market segment.Because the data points from bull markets and bear markets are separated,the regression calculation of each bull market or bear market will not be affected by the invaders originated from other market segments.This improves the relevancy of the regression results.Moreover,because the target betas of the invaders in a certain bull or bear market are the same as those of the aboriginals,they are suitable for use in the regression calculation together with the aboriginals.The reliability of linear regression results increases with larger span of the data points in the direction of the regression line.An invader is a data point extending from the side of the aboriginals to the opposite side.Therefore,the span made by the invaders and aboriginals together would be significantly larger than that made by aboriginals alone.Compared with the fine-tuned versions of the Treynor model and Henriksson model obtained by removing the invaders from the data sets,by including the invaders in the regression calculation,the Bull and Bear model tends to give regression results which are closer to the true values.In short,for traditional models,the invaders are troublemakers.Raising the frequency of excess return sampling will worsen the damage caused by invaders to the validity of the regression results.For the Bull and Bear model,invaders are contributors to the precision of the regression results.Four types of simulation funds were designed to compare the effectiveness of the Bull and Bear Model with those of the traditional models in the market represented by the S&P 500 index in the past eighteen years.The four types of simulation funds represent respectively four types of fund managers showing different timing styles:those taking the timing move exactly at the moment the long-term market trend reverse,those always taking the timing move several days ahead,those always taking the timing move several days after,and those spreading out the timing move evenly over a period extending several days before to several days after the reversal of the long-term market trend.The Bull and Bear Model is better than the traditional models and their finetuned versions in identifying timing abilities in the above four types of simulation funds,which demonstrated perfect or near-perfect timing performances.After that,from the database provided by Morningstar Corporation,about 27,000 open-ended mutual funds investing in the United States market were searched using the keyword “balanced” in their names.In the search results,the funds failing to cover the entire period from March 27,2000 to January 26,2018 were removed.259 balanced funds remained and were assessed using the above five models/versions.Balanced funds include timing maneuvers as one of their investment strategies so there are reasons to expect better chance to encounter funds showing timing ability among them.The results showed that the Bull and Bear model reported the most number of cases with timing abilities.Unlike the case with the above four types of simulation funds,there is no way to know the actual magnitude of change in the beta of an assessed fund.Thus one cannot draw the conclusion that higher scores must imply better assessment model,as in the simulation fund cases.By visually inspecting the funds each of the five models/versions recommended the most,it was found that recommendation made by the Bull and Bear model was the most trustworthy.Finally,the timing abilities of Hong Kong MPF managers were assessed using the Bull and Bear model.The results reported by the model are similar to those reported by the traditional models.Conclusion 1: Mandatory provident fund managers in Hong Kong have generally shown no timing ability.Conclusion 2: The assessment model designed for the timing ability of long-termreturn oriented fund managers is more effective than traditional models when daily returns are used.This article achieves its first innovation by noticing that the rule assumed by the traditional models for guiding the timing actions of the managers with prefect timing abilities is return-data-sampling-frequency-oriented in nature,thus the timing styles these models aim at will shift with changes in return data sampling frequency.With low return data sampling frequency the models aim in effect at managers with longterm-return-oriented timing abilities,while with high return data sampling frequency the models aim in effect at managers with short-term-return-oriented timing abilities.Increasing return data sampling frequency will make sampling-period-return-oriented become short-term-return-oriented in effect.When this idea is translated into operation,it turns out that the increase in return data sampling frequency will widen the difference between the graphs of a long-term-return-oriented manager and a sampling-periodreturn-oriented one,both with perfect timing abilities,on the excess investment portfolio return vs.excess market portfolio return coordinate plane.The difference is large enough to prevent the traditional models which are sampling-period-returnoriented from making a fair assessment on the timing ability of long-term-returnoriented fund managers using high return data sampling frequency.This article achieves its second innovation by proposing a fine-tuned version for each of the traditional models in order to enhance their effectiveness in assessing the timing ability of long-term-return oriented fund managers with high sampling frequency return data.This article achieves its third innovation by,to follow the trend of promoting data sampling frequency,designing a new assessment model to meet the needs of assessing the timing ability of long-term-return oriented fund managers while the traditional models aim at short-term-oriented fund managers.The new model does not rely on adding conditions to the data used by the tradition model to improve the statistical reliability of the results to work,it is obtained by making a change to one of the cornerstones of the traditional models,the criteria for setting high or low betas.This article achieves its fourth innovation by broadening the idea of timing from a fixed point into a spectrum,and thus establishing new positioning for the above five models / versions for evaluating fund managers' timing capability in the context of high data sampling frequency.The two traditional models are responsible for assessing short-term-return oriented timing abilities.The generalized new model is responsible for assessing long-term-return oriented timing abilities.Cases in between are to be handled by the generalized fine-tuned versions of the two traditional models.Not only applicable to MPF in Hong Kong,the three new model/versions proposed in this paper are also applicable to the timing ability assessment of long-termreturn oriented fund managers of all open-ended funds in any market in the world.Improvement 1 awaiting: Based on the design characteristics of this model,it can only compare the timing abilities among fund managers who are long-term-return oriented.For a fund manager who makes timing choice based on short-term returns,his/her well founded speculative timing maneuvers will be incorrectly interpreted as failing long-term-return oriented moves.However,fund managers in real life may make timing moves with varying degrees of speculation in the management of active funds.If some tolerance on speculative moves is added to an assessment model which mainly targets the timing ability of long-term-return oriented fund managers,the practicality of the entire mechanism will be enhanced.Improvement 2 awaiting: The extent the fund under assessment deviates from a fund showing perfect timing ability is all summed up in one single statistical parameter,i.e.,the adjustment amount of the fund's beta in the favorable direction.This figure alone cannot tell whether the fund being assessed is making timing moves ahead of the reversal of long-term market trend,or showing lags in actions,or acting in some other styles.If improvements can be made to the new model in this aspect,its quality will be promoted to a higher level because of the enhanced details of the assessed it can depict. |
PDF Full Text Request