Aggregation Issues in Operational Risk
|
|
- Cody Melton
- 6 years ago
- Views:
Transcription
1 Aggregation Issues in Operational Ris Rosella Giacoetti Departent of Matheatics, Statistics, Coputer Science and Applications L. Mascheroni School of Econoics and Business Adinistration, University of Bergao Via dei Caniana, 4 Bergao, Italy E-ail: rosella.giacoetti@unibg.it Svetlozar Rachev Departent of Econoetrics, Statistics and Matheatical Finance School of Econoics and Business Engineering, University of Karlsruhe Postfach 6980, 768 Karlsruhe, Gerany and Departent of Statistics and Applied Probability, University of California, Santa Barbara CA , USA E-ail: rachev@statisti.uni-arlsruhe.de Anna Chernobai Departent of Finance, M.J. Whitan School of Manageent, Syracuse University 7 University Ave, Syracuse, NY ,USA E-ail: annac@syr.edu Marida Bertocchi Departent of Matheatics, Statistics, Coputer Science and Applications L. Mascheroni School of Econoics and Business Adinistration, University of Bergao Via dei Caniana, 4 Bergao, Italy E-ail: arida.bertocchi@unibg.it Aggregation Issues in Operational Ris
2 Abstract In this paper we study copula-based odels for aggregation of operational ris capital across business lines in a ban. A coonly used ethod of suation of the value-at-ris (VaR easures, that relies on a hypothesis of full correlation of losses, becoes inappropriate in the presence of dependence between business lines and ay lead to over-estiation of the capital charge. The proble can be further aggravated by the persistence of heavy tails in operational loss data; in soe cases, the subadditivity property of value-at-ris ay fail and the capital charge becoes underestiated. We use α-stable heavy-tailed distributions to odel the loss data and then apply the copula approach in which the arginal distributions are consolidated in the syetric and sewed Student t- copula fraewor. In our epirical study, we copare VaR and conditional VaR estiates with those obtained under the full correlation assuption. Our results deonstrate significant reduction in capital when a t-copula is eployed. However, the capital reduction is significantly saller than in cases where a oderately heavy-tailed or thin-tailed distribution is calibrated to loss data. We also show that for confidence levels below 94% VaR exhibits the super-additivity property.
3 . INTRODUCTION The financial industry s attention to operational ris has been on increase in recent years. Operational ris is defined as the ris of loss resulting fro inadequate or failed internal processes, people and systes or fro external events. Operational ris is believed to be largely a fir-specific non-systeatic ris: According to the Basel Coittee, unlie aret and perhaps credit ris, the [operational] ris factors are largely internal to the ban. In 00, the Basel II Capital Accord (hereforth, Basel II provided a detailed set of guidelines for bans on the basis of which they are required to estiate operational ris capital to serve as a buffer against potential future losses. The fir-specific nature of operational ris has propted developent of statistical odels that ae efficient use of historic operational loss data on the basis of which the capital charge can be estiated. The ost sophisticated of such approaches, and the one ost favoured by regulators 3 is the Loss Distribution Approach (LDA. LDA falls in the category of the Advanced Measureent Approaches. 4 LDA is the ost accurate fro the statistical point of view as it utilizes the exact distribution of historic losses both frequency and severity and is based on an individual ban s internal loss data. The core principle of the capital charge estiation under this approach is the value-at-ris (hereforth, VaR etric that is easured based on a principle of aggregation of the frequency and severity distributions of losses forecasted for a one-year ahead tie horizon. Two tass are central to an accurate estiation of operational VaR. The shape of the upper tail of the loss distribution largely deterines the aount of the ris capital. One approach, extree value theory (EVT, involves a separation of the ain body of the loss distribution fro the tails and odelling the tails with a Generalized Pareto Distribution (GPD; see for exaple, Chavez-Deoulin, Ebrechts, and Neslehova (006. In operational ris, EVT has been also used to odel external data that is then used to populate scarce internal data; see Baud, Frachot, and Roncalli (00 for a ore detailed discussion. 5 Aggregation of operational losses across business units or event types (or both reains as iportant issue. A siplistic approach involves estiation of VaR easures for each cell independently and the adding the up to produce the aggregate easure of ban s ris. The proble with such an approach is that it assues a perfect positive correlation between cells. As a result, the aggregate easure of ris represents an upper bound for a ban s true total level of ris. This property, often referred to as sub-additivity of ris easures, suggests that a ban can effectively reduce its ris capital by taing into account the dependence structure that exists between cells. See BCBS (00b. See BCBS ( Bans are allowed to choose an approach based on the ban s size and ris exposure and the ability to eet required criteria. 4 A detailed description of the approaches to easure the operational ris capital charge is docuented in BCBS (00a,b; Overview of EVT and its applications to ris anageent can be found in Ebrechts, Kluppelberg, and Miosch (997. Applications to operational ris are discussed in Chernobai, Rachev, and Fabozzi (007 and McNeil, Frey, and Ebrechts (005, aong others. 3
4 Soe probles with sub-additivity are often cited in literature. The sub-additivity property ay fail when ris is easured by VaR (see Artzner, Delbaen, Eber, and Heath (999. In this sense, VaR is not a coherent easure of ris. On the contrary, conditional value-at-ris (CVaR is coherent in that the sub-additivity property holds. The failure of sub-additivity (i.e., super-additivity ay be aggravated in the presence of heavy tails in the loss data. As shown in Ibragiov and Walden (007, this ay have adverse iplication for diversification. Our epirical analysis, presented later in this paper, also shows evidence of super-additivity. In this paper, we extend discussion in Giacoetti, Rachev, Chernobai, Bertocchi, and Consigli (007 that focused on various odelling techniques for operational loss data. Here, we exaine the issue of aggregation of operational losses across different business lines using a copula approach. Copulas are gaining popularity in easuring dependence in financial ris. They will be described later in this paper. In operational ris, copulas have been applied by Chavez-Deoulin, Ebrechts, and Neslehova (006, Ebrechts and Puccetti (006a,b, Dalla Valle, Fantazzini, and Giudici (007, and Chapelle, Craa, Hubner, and Peters (004, to nae a few. Using copulas tends to lead to significant reductions in total VaR. For exaple, Chapelle, Craa, Hubner, and Peters (004 showed that for a 99% VaR a ban can achieve a 35% reduction in capital estiates by using copula, echoing the results in Dalla Valle, Fantazzini, and Giudici (007 who deonstrated that using copulas can result in savings for a ban in the range of 30% to 50%. We use t-copulas to odel the dependence between losses. t-copulas are optial for odelling dependence between operational losses fro different groups in that they succeed in capturing dependence in the tails. Other copulas, such as the Gaussian copula, has no tail dependence. For the loss distribution we use (i a variation of the α-stable distribution and (ii a ixture of the α-stable distribution and Generalized Pareto distribution. Such choice of loss distributions ensures that we do not leave out any extree events and they are appropriately accounted for. We apply the ethodology in an epirical study to operational loss data of a European ban. The paper is organized as follows. Section describes a statistical odel for operational ris. Section 3 presents the data and epirical ethodology. Section 4 describes the results of applying copulas to operational loss data and its iplications for operational ris anageent. Finally, Section 5 concludes the paper and suarizes the findings.. A Modified Loss Distribution Approach for Operational Ris The concept of copound Poisson process provides an accurate analytical fraewor to address the odelling proble of operational ris and is utilized in the Loss Distribution Approach (LDA of the Advanced Measureent Approaches proposed by Basel II. The tiing of the events is captured by the intensity of the Poisson process and the losses by an appropriate state distribution. Consider a ban with K business lines/event type 4
5 cobinations, i=,,,k. Then the aggregate operational loss for i-th business line are i considered to follow a rando process L t with t 0 i N t i i i i Lt X, X F, Nt Poi(. ( 0 The operational loss distribution is thus jointly deterined by the average nuber of losses per unit of tie the intensity of the Poisson process N t, the counting process with integer variables and by the loss agnitudes X in onetary ters observed over a pre-deterined interval of tie usually taen to be one year. In our case X belong to a faily F of paraetric continuous distributions. Value-at-Ris is a ris easure that can be used as proxy for capital charge. It is coputed as a high quantile (such as 99.9% of the aggregate loss distribution. For cell i, i VaR G (0.999 ( t L Where G is the cuulative distribution of L. Then, under the LDA, the total capital charge for a ban with M cells can be estiated as the su of VaR easures 6 across cells: K M i i VaR t This LDA ethodology assues that the losses belonging to different business line/event type cells are perfectly positively correlated with each other. However, if the correlation is not perfect, under nice conditions LDA provides an upper bound for the total capital charge for a ban. As a result, LDA would generally tend to over-estiate the aount of ris capital a ban should use. (3 3. Data and Methodology The data set selected for the study covers operational losses for a large European ban fro January 00 to August 005 a total of 3 years and 8 onths. The saple consists of a total of slightly under,700 observations. The data are classified in accordance with the Basel II guidelines into business lines and event types. Due to relatively sall data saples, the business lines considered for this study are: Retail Baning (using the Basel II definitions, this refers to business line 3 or BL3, Coercial Baning (BL4, and Retail Broerage (BL8. We do not further granulate the data by event types for our estiations. BL3 accounts for 77.69% of the data, BL4 accounts for 7.8%, and BL8 accounts for 8.7% of our saple. 6 An alternative to VaR is conditional VaR (or CVaR which is estiated as the average loss given that it exceeds VaR. 5
6 Table. Descriptive statistics for internal, external, and pooled operational loss data (in Euro. Business Line Retail Baning (BL3 Coercial Baning (BL4 Retail Broerage (BL8 Internal Data External Data Pooled Data Saple Statistic Logscale original Log-scale original Log-scale original Mean 5, , , Median, , , St. Dev. 97, , , Sewness Kurtosis Mean 8, , , Median, , , St. Dev. 33, , , Sewness Kurtosis Mean 3, , , Median, , , St. Dev. 48, , , Sewness Kurtosis , , We populate our saple with additional data extracted fro an external database. The external database is a consortiu-type database provided by a European vendor. 7 The data that constitute the database are collected fro nearly 00 institutions. The size of the external data used for the analysis in this paper is approxiately six ties the saple size of internal data. In the external database, BL3 accounts for 60.63%, BL4 accounts for 6.46%, and BL8 accounts for 8.7% of the data. To cobine the internal data with external, we first standardized all data by the respective eans and standard deviations and then scaled the pooled data bac using the ean and the standard deviation of the original internal data to obtain an expanded internal dataset. Table suarizes descriptive statistics of the data. For the frequency distributions we considered a non-hoogeneous Poisson process odel; see Giacoetti et al. (007 for further details. For the loss distributions, we considered a large spectru of candidate distributions: Lognoral, Generalized Pareto, Weibull, Logweibull, and syetrised α-stable. 7 We refrain fro providing the consortiu nae in order to preserve confidentiality of the ban. 6
7 Syetrised α-stable distribution belongs to the class of stable distributions. 8 For the syetrised α-stable distribution we syetrised the data by applying the transforation Y = [X; -X] to the original data X and then fitting a -paraeter (shape and dispersion syetric α-stable distribution. The advantage of fitting a syetric α- stable distribution to syetrised data over fitting a 4-paraeter (shape, dispersion, sewness, and location α-stable distribution to original data is that syetric α-stable distribution requires estiation of only two paraeters (shape and dispersion allowing for ore efficient estiates. The shape paraeter α identifies the heaviness in the tail: α> refers to a oderately heavy tail with a finite ean, while α indicates a very heavy-tailed distribution with an infinite ean. We refer as Model to the approach in which a hypothesized loss distribution is fitted to the entire dataset. Goodness-of-fit test results showed that the syetric α-stable distribution resulted in best fit to our full-saple loss data. 9 For the three business lines, BL3, BL4, and BL8, the shape paraeters were estiated as 0.99, 0.83, and 0.99, respectively, and point to a very heavy tail. Table. Estiates of the shape paraeters. Panel A: Full-saple distribution approach (Model α of the syetrised α-stable distribution BL BL BL Panel B: Mixture distribution approach (Model Body α of the syetrised α-stable distribution Tail /ξ of the Generalized Pareto distribution MLE estiate Hill estiate BL BL BL In a separate approach (Model, we cobined Extree Value Theory with the notion of ixture distributions to odel the loss data. We first identified a level of threshold beyond which the losses were assued to follow the Generalized Pareto Distribution (GPD. We then estiated the paraeters of the GPD for the tails with the Maxiu 8 See Rachev and Mittni (000 for a thorough discussion. 9 See Giacoetti et al. (007 for a detailed description of the goodness-of-fit test results. 7
8 Lielihood estiator (MLE and the Hill estiator, and used syetric α-stable distribution to odel the body of the loss data. Finally, we cobined the two distributions into a -odel ixture distribution constructed as the weighted average of the two eber distributions. Table suarizes the estiated shape paraeters fro Models and. The estiates suggest very heavy-tailed ixture distributions especially for BL4. 4. Aggregation of Losses Using Copulas This section focuses on the discussion of two issues. The first issue is related to dependence between losses belonging to different business line/event type cobinations. 0 When the correlation is sub-perfect, i.e. Equation 3 provides an upper bound for a ban s ris capital, the nowledge on the for of the dependence structure is needed. Coputing correlation between cells iplies a linear for of dependence and therefore ay not be an optial solution. If the dependence is of a non-linear nature, then copulas provide a natural solution. A second issue we will address in this section is exaining scenarios in which subadditivity property of VaR fails. Sub-additivity property dictates that i VaR i VaR Li (4 i Sub-additivity property ay fail in cases when a loss distribution is very heavy-tailed. In essence, this eans that the VaR of aggregate losses ay actually exceed the su of the VaR easures coputed for each cell separately. 4.. Definition of Copula Definition (Copula: A copula function is a apping fro a set of univariate arginals to their full ultivariate distribution. For unifor rando variables U, U,..., U, the joint distribution function C, or copula, is C ( u, u,..., u, P U u, U u,..., U u. (5 Copula functions can be used to lin given arginal distributions with a joint distribution, since for given arginal distribution functions F ( l, F ( l,..., F ( l, we have: C( F ( l,..., F ( l, P U F ( l,..., U F ( l P P F ( U l,..., F ( U L l,..., L l F( l,..., l l (6 0 For the sae of siplicity and data consideration, in this paper we classify losses by business lines only. 8
9 Slar (959 established the fundaental converse result: he showed that any ultivariate distribution function F can be written in the for of a copula function, naely: If F l, l,..., l is a joint ultivariate distribution function with univariate ( arginal distribution functions F l, F ( l,..., F ( l ( ( l, l,..., l C( F ( l, F ( l,..., F ( l, then there exists a copula function C( u, u,..., u such that F. If each Fi is continuous then the copula is unique. Although copulas ay be difficult to wor with, the convenient aspect of copula estiation is that it can be perfored independently fro the estiation of the arginal severity distributions. In this paper, we consider syetric Student t-copula and a sewed Student t-copula because these copulas are capable of capturing the dependence in the upper tail of the distributions. 4.. Syetric Student t-copula A syetric Student t-copula c( u,..., u, R, is a function of the degrees of freedo and the correlation atrix R. For the 3-business line exaple, the syetric t-copula has joint distribution function of the for: 3 T ( 3 / ( 3 / ( / ( w R w 3 ( / ( / ( / ( w / c ( u, u, u3; R, (7 0.5 R i,,3 T where the vector w ( w, w, w, wi t ( ui has coponents directly coputed fro the inverse t distribution. The degrees of freedo of the copula can be estiated using a recursive procedure proposed by Mashal and Naldi (00. Given the initial estiate T 3 T i ~ ~, (,,, i i i R0 zt zt zt zt zt zt zt ( lt, lt F( lt (8 T t,.., T the degrees of freedo in Equation 7 are deterined as optial values of the log-copula: For increasing j j d, 0, we generate recursively for 0,,,... 3 T T j tttt 3 ~ ~ ~ 3 ~ i i R tt : ( tt, tt, tt t ( lt, lt, lt, lt F( lt (9 T t j T tt Rtt j Then, a liit correlation atrix is input into the copula function up to the point in which the log-lielihood is axiised for the current correlation atrix and degrees of freedo: T t 3 ~ ~ arg ax ln c ~ ( l, l, l, R, R li ( R. (0 t, R t t t j,,..., j j The iterative procedure will converge to a copula estiate that can be then incorporated into the siulator. i 9
10 We estiated the liit value of the degrees of freedo to be between 6 and 7. The degrees of freedo and the optial correlation atrix allow for a correct definition of the ultivariate density for a ban s aggregate losses. The epirical correlation and copula-based correlation structures can be illustrated in a 3- diensional space. Figure shows the joint losses of the three business lines internal data on the top left and the epirical joint distribution on the top right. Siulated losses are on the botto left and the syetric t-copula-based joint distribution are in the botto right corner. Coparison of the figures in the top row and in the botto row suggests that the syetric t-copula succeeds in effectively reproducing the true dependence structure between the losses. Figure. Correlation structure of the losses in the three business lines and a canonical exaple with a syetric t-copula. Top left: Weely aggregated losses, epirical data; top right: Epirical distribution of weely aggregated losses; botto left: Weely aggregated losses, siulated data; and botto right: Distribution of siulated losses, weely aggregated using syetric t-copula. 0
11 4.3. Sewed Student t-copula We use the following for of the ultivariate sewed Student t-distribution for the copula function, defined by the stochastic representation as follows: X : W Z W ( where W IG(, and Z N(0,, Z is independent of W, (,, n is an n- diensional vector accounting for the sewness, (,, n is n-diensional location paraeter vector, and ν is the degrees of freedo. We denote this distribution by X t n (,,. The notation IG(, stands for the inverse Gaa distribution with both paraeters equal to. Thus, W is a one-diensional rando variable and Z is a rando vector having a zero-ean ultivariate noral distribution with covariance atrix n n. ( n n nn The ultivariate sewed Student s t-distribution allows for closed-for expression of its density, f X ( x ak ( n ( ( ( x ( ( x ( x ( x n 4 exp(( x ( x ( ( x n n (3 where n x R, K is the odified Bessel function of the third ind and n a ( ( n. (4 The sewed Student s t copula is defined as the copula of the ultivariate distribution of X. Therefore, the copula function is C( u, u,..., un FX ( F ( u,, Fn ( un (5 For ore inforation, see Section.7 in Rachev and Mittni (000 and Dearta and McNeil (005.
12 where FX is the ultivariate distribution function of X and F ( u,,, n, is the inverse of the distribution function of the -th arginal of X. That is, F X (x has the density defined in Equation 3 and the density function f (x of each arginal is f ( x ak ( n ( ( x ( ( x (( 4 exp(( x ( x (, x R (6 where is the -th diagonal eleent in the covariance atrix. defined in Equation. For the sewed Student s t-copula estiation we assue that there are n business lines/event types cobinations of aggregate loss data saples: X X, X,, X T n ( n. We first estiate the paraeters of the sewed Student s t-ultivariate distribution on historical operational losses following the following procedure fro Rachev, Stoyanov, and Milov (007: Step. Fit one-diensional sewed Student s t-distribution over all ris variables on a stand-alone basis. The result fro that step is: ( 5, ˆ, ˆ, ˆ, i, n (7 i i i, We use MLE ethod to obtain the estiates. For the degrees of freedo we use γ=5 because at this value the copula is ost sensitive to the asyetry of paraeters; the sewed distribution reduces to the syetric case at 0. Step. Estiate the correlation atrix of the ultivariate sewed Student s t distribution by the following forula: ˆ (cov( X ˆˆ (8 ( ( 4 where 5 and ( ˆ, ˆ,, ˆ. ˆ n Step 3. Adjust the atrix ˆ so that it becoes positive definite. Having estiated the paraeters of the sewed t-distribution, we obtain the sewed t- copula using the following -step siulation algorith: Algorith Step. Draw N independent n-diensional vectors fro the ultivariate sewed Student s t distribution using the stochastic representation defined in Equation and the set of fitted paraeters ( 5, ˆ i, ˆ ˆ i, i, i,, n. The result fro that step is N n atrix S s ij with siulations. This is obtained by, first, drawing N independent n-diensional vectors fro the ultivariate Noral distribution N (0,, second, drawing N independent rando
13 nubers fro the inverse Gaa distribution with paraeters IG (,, and, third, obtaining final siulations using Equation with the estiated paraeter values. Algorith Step. Transfor siulations S to unifor siulations U using the saple distribution function of the arginals. Denote by Fˆ ( x the saple cuulative distribution function of the -th arginal, j N F ˆ ( x I S j x N, (9 where I{A} stands for the indicator function of the set A. Then U Fˆ ( S, j,, N, n. (0 j j, 4.4. Results of Copula Estiation with Operational Loss Data In this section, we apply the syetric Student t-copula and sewed Student t-copula to our data saples. We consider two odels for the arginal distributions of the losses that were suarized in Section 3. Table 3 suarizes the estiates of population descriptive statistics and ris capital easures at confidence levels 97.5, 98, 99, and 99.9 percent, by business line. These statistics refer to the arginals that will later be aggregated with copulas. It is notable that Model produces uniforly significantly larger estiates than Model. In order to estiate concordance easures between the three business lines, we aggregated losses on a weely basis. Table 3 presents results of the Spearan correlation coefficient and Kendall correlation coefficient for each pair of the business lines. The estiates suggest very low degree of dependence between the losses. To aggregate the losses belonging to the three business lines, we use the syetric Student t-copula and sewed Student t-copula, described in Sections 4. and 4.3, respectively. The degrees of freedo were estiated for the forer copula as ν=6.7 and for the latter copula as ν=5. To apply the copulas to the loss data, we first estiate the copula paraeters using historical losses aggregated weely. We then repeat the following steps a very large nuber of ties. In the first step, fro the estiated copula we saple a ultivariate rando vector with arginals distributed as unifor [0,] rando variables. In the next step, for each business line, we obtain scenarios for the cuulative loss realization by inverting the unifor [0,] variate fro the previous step and then su the up to obtain the total loss for the ban. Repeating these steps produces the distribution of aggregate losses. Finally, VaR and CVaR are coputed fro the obtained aggregate distributions. 3
14 Table 3. Population descriptive statistics and ris capital estiates by business line. Panel A: Model. Population descriptive statistics (Euro Ris capital estiates (Euro VaR CVaR Mean St.Dev. 97.5% 98% 99% 99.9% 97.5% 98% 99% 99.9% BL BL BL Panel B: Model. Population descriptive statistics (Euro Ris capital estiates (Euro VaR CVaR Mean St.Dev. 97.5% 98% 99% 99.9% 97.5% 98% 99% 99.9% BL BL BL Table 4. Estiates of concordance. BL3 Spearan Kendall BL4 Spearan Kendall BL8 Spearan Kendall BL3 BL4 BL
15 Table 5. Population descriptive statistics and ris capital estiates for aggregate loss data. The nubers in parentheses indicate percentage reduction (if - or increase (if + fro the corresponding ris easure under perfect positive correlation scenario. Panel A: Model. Perfect positive correlation Syetric t-copula Sewed t-copula Population descriptive statistics (Euro Ris capital estiates (Euro VaR CVaR EL UL 97.5% 98% 99% 99.9% 97.5% 98% 99% 99.9% (-.8% (-.8% 0.77 (-.% 0.98 (+6.4% 0.87 (-3.0% 0.8 (-5.4% (-5.5% (-.9% (-9.8% (-3.7% (-6.4% (-5.%(-38.4% (+.% (-3.7% (-5.7% (-3.% (-3.7% (-.9%(-34.% Panel B: Model. Perfect positive correlation Syetric t-copula Sewed t-copula Population descriptive statistics (Euro Ris capital estiates (Euro VaR CVaR EL UL 97.5% 98% 99% 99.9% 97.5% 98% 99% 99.9% (-.9% 0.48 (-4.5% (-.6%.004 (0.0% Panel C: Historical data. Population descriptive statistics (Euro (-0.7% (-.8%.0 (+3.%.00 (+3.0%.936 (-.4%.857 (-5.4% (-0.8% (-3.7% (-4.7% (-3.0% (-37.4% (-8.% (-33.3% (-6.8% (-4.6% (-35.% Ris capital estiates (Euro VaR CVaR EL UL 97.5% 98% 99% 99.9% 97.5% 98% 99% 99.9% Historical data Historical data excluding worst losses
16 Table 5 suarizes the estiates for expected and unexpected losses (EL and UL and the ban s cuulative ris capital, derived fro the copula-based approach and copared to the perfect correlation approach. The estiates are based on a one wee horizon. It is notable that in our results there is a negligible difference in EL and UL and the effect is ore pronounced for the VaR and CVaR estiates. In vast ajority of cases copula approach results in a substantial reduction in ris capital for the ban. For exaple, for 99.9% VaR, reduction in capital ranges fro % to 9% and for 99.9% CVaR, reduction in capital is roughly in the 34%-38% range. Another notable result is that if Model is used, syetric Student t-copula generally produces higher reduction in ris capital than the sewed Student t-copula, while the relation is reverse in ost part if Model is used. However, because sewed Student t-copula is a generalized version of the syetric Student t-copula, a ris anager would prefer the forer one. Panel C of Table 5 shows capital estiates based on historical data. Coparison of the figures with those under Model and Model reveals that for percentiles below 99.9, the forer odel uniforly under-estiates the true historic loss while the latter odel produces results fairly consistent (only slightly higher with those actually experienced. Then, because Model corresponds to a ore heavy-tailed distribution than Model, the CVaR figures under Model reveal overestiation of the historic counterparts roughly by a factor of. On the contrary, Model in ost part underestiates CVaR. A ris anager would favour Model over Model based on the above discussion. Our findings indicate lower reduction in ris capital fro those reported by Chapelle, Craa, Hubner, and Peters (004 and Dalla Valle, Fantazzini, and Giudici (007. Chapelle, Craa, Hubner, and Peters (004 showed that for a 99% VaR a ban can achieve reduction in capital estiates by 35% by using copula, and Dalla Valle, Fantazzini, and Giudici (007 showed reduction in the range of 30%-50%. One explanation is that we used a uch ore heavy-tailed loss distribution than in the two other studies. This eans that, if a thin-tailed loss distribution is used to odel loss data, copula-based aggregation would result in uch ore significant reduction in total capital. In essence, this iplies that if a ban has istaenly chosen to use a thin-tailed loss distribution when the data are in fact heavy-tailed, the resulting capital would be understated. The converse is also true. The choice of the loss distribution thus becoes of central concern and ust be estiated with high degree of accuracy. 6
17 VaR( L i - (VaR i % 8.0% 8.40% 83.60% 84.80% 86.00% 87.0% 88.40% 89.60% 90.80% 9.00% 93.0% 94.40% 95.60% 96.80% 98.00% 99.0% Euro Confidence level Figure. Illustration of super-additivity and sub-additivity of historic VaR Super-Additivity of VaR Panels A and B of Table 5 show that for the 98% confidence level, we observe superadditivity of VaR easures in 3 out of 4 cases. Super-additivity is also observed in our results for the historic estiates of VaR under 99.9% for Models and and CVaR estiates under 99.9% for Model. Super-additivity has been docuented in literature; see, for exaple, McNeil, Frey, and Ebrechts (005. Figure illustrates superadditivity and sub-additivity in historical VaR estiates. The horizontal line is the benchar representing a VaR estiate based on the aggregation of historical data for BL3, BL4, and BL8 using historical correlations. The plot represents the suation of individual VaR estiates for each of the three business lines. Super-additivity is observed for confidence levels under 94%. One ust pay extra care when choosing a confidence level to provide VaR and CVaR easures: for confidence levels chosen too low, the estiated capital charge ay be under-estiated while for confidence levels too high, it ay be over-estiated. The phenoenon of super and sub-additivity sees ore severe for ore heavy-tailed loss distributions. 7
18 5. Conclusions In this epirical paper we used a copula approach to odel the dependence between operational losses belonging to different business lines. The contributions of this paper can be suarized as follows: ( For the loss distributions that constitute the arginals of copulas we used very heavy-tailed syetric α-stable distribution (a eber of α-stable distributions and a cobination of a syetric α-stable distribution with the Generalized Pareto distribution for the tails. In operational ris, such heavy-tailed distributions are superior to thin-tailed and oderately heavy-tailed distribution due to their capacity to detect and account for the heaviness in the upper tails. ( We selected syetric Student t-copula and sewed Student t-copula due to their ability to capture tail dependence. Our epirical study has shown that using t-copulas results in substantial reduction in ris capital for the ban. Although our findings are consistent with other epirical studies that have docuented capital reduction when using copula, upon drawing coparisons with other siilar studies, we have found that the agnitude in capital reduction is saller than if thin-tailed or oderatelytailed distributions, such as Lognoral and Gaa, are used to constitute the arginals. (3 Another finding was presence of the super-additivity phenoenon in the VaR estiates. For our data saple, super-additivity was observed for confidence levels below 94%. Our findings effectively deonstrate that falsely settling on a loss distribution that is not sufficiently heavy-tailed for given data puts the capital estiates at the ris of severe underestiation for very high confidence levels or overestiation for confidence levels insufficiently high. Careful analysis and a variety of goodness-of-fit tests are thus crucial in selecting a loss distribution which is central to the accuracy of the capital charge estiates. References Artzner, P., Delbaen, F., Eber, J.-M., and Heath, D. (999, Coherent easures of ris, Matheatical finance, 9(3, pp Baud, N., Frachot, A., and Roncalli, T. (00, Internal data, external data, and consortiu data for operational ris easureent: How to pool data properly, technical report, Groupe de Recherche Operationnelle, Credit Lyonnais, France. BCBS (998, Operational ris anageent, Ban of International Settleents, Basel, Switzerland, BCBS (00a, Consultative docuent: Operational ris, Ban of International Settleents, Basel, Switzerland, 8
19 BCBS (00b, Woring paper on regulatory treatent of operational ris, Ban of International Settleents, Basel, Switzerland, BCBS (006, International convergence of capital easureent and capital standards, Ban of International Settleents, Basel, Switzerland, Chernobai, A., Rachev, S. T., and Fabozzi, F. (007, Operational ris: A guide to Basel II capital requireents, odels, and analysis, Wiley Finance Series, John Wiley & Sons, Hoboen, New Jersey. Dearta, S. and McNeil, A. (005. The t copula and related copulas, International Statistical Review, 73(, -9. Ebrechts, P., Kluppelberg, C., and Miosch, T. (997, Modeling extreal events for insurance and finance, Springer-Verlag, Berlin. Ebrechts, P. and Puccetti, G. (006a, Aggregating ris capital, with an application to operational ris, The Geneva Ris and Insurance Review, 3(, pp Ebrechts, P. and Puccetti, G. (006b, Bounds for functions of dependent riss, Finance and Stochastics, 0, pp Giacoetti, R., Rachev, S. T., Chernobai, A., Bertocchi, M., and Consigli, G. (007, Heavy-tailed distributional odel for operational ris, The Journal of Operational Ris, (, pp Ibragiov, R. and Walden, J. (007, The liits of diversification when losses ay be large, Journal of Baning and Finance, 3, pp Mashal, R. and Naldi, M. (00, Pricing ulti-nae credit derivatives: Heavy tailed hybrid approach, Woring paper, Colubia Business School. McNeil, A. J., Frey, R., and Ebrechts, P. (005, Quantitative ris anageent: Concepts, techniques, and tools, Princeton Series in Finance, Springer, Boston. Rachev, S. T. and Mittni, S. (000, Stable Paretian odels in finance, John Wiley & Sons, New Yor. Rachev, S. T., Stoyanov, S., and Milov, G. (007, Sewed Student t Copula, Technical report, FinAnalytica Inc. Slar, A., (959, Fonctions de répartition à n diensions et leurs arges, Publications de l'institut de Statistique de L'Université de Paris, 8, pp
III. Valuation Framework for CDS options
III. Valuation Fraework for CDS options In siulation, the underlying asset price is the ost iportant variable. The suitable dynaics is selected to describe the underlying spreads. The relevant paraeters
More information... About Higher Moments
WHAT PRACTITIONERS NEED TO KNOW...... About Higher Moents Mark P. Kritzan In financial analysis, a return distribution is coonly described by its expected return and standard deviation. For exaple, the
More informationFinancial Risk: Credit Risk, Lecture 1
Financial Risk: Credit Risk, Lecture 1 Alexander Herbertsson Centre For Finance/Departent of Econoics School of Econoics, Business and Law, University of Gothenburg E-ail: alexander.herbertsson@cff.gu.se
More informationAnalysis of the purchase option of computers
Analysis of the of coputers N. Ahituv and I. Borovits Faculty of Manageent, The Leon Recanati Graduate School of Business Adinistration, Tel-Aviv University, University Capus, Raat-Aviv, Tel-Aviv, Israel
More informationCalculating Value-at-Risk Using the Granularity Adjustment Method in the Portfolio Credit Risk Model with Random Loss Given Default
Journal of Econoics and Manageent, 016, Vol. 1, No., 157-176 Calculating Value-at-Risk Using the Granularity Adjustent Method in the Portfolio Credit Risk Model with Rando Loss Given Default Yi-Ping Chang
More informationTime Value of Money. Financial Mathematics for Actuaries Downloaded from by on 01/12/18. For personal use only.
Interest Accuulation and Tie Value of Money Fro tie to tie we are faced with probles of aking financial decisions. These ay involve anything fro borrowing a loan fro a bank to purchase a house or a car;
More informationEstimating Nonlinear Models With Multiply Imputed Data
Estiating onlinear Models With Multiply Iputed Data Catherine Phillips Montalto 1 and Yoonkyung Yuh 2 Repeated-iputation inference (RII) techniques for estiating nonlinear odels with ultiply iputed data
More informationAn alternative route to performance hypothesis testing Received (in revised form): 7th November, 2003
An alternative route to perforance hypothesis testing Received (in revised for): 7th Noveber, 3 Bernd Scherer heads Research for Deutsche Asset Manageent in Europe. Before joining Deutsche, he worked at
More informationPRODUCTION COSTS MANAGEMENT BY MEANS OF INDIRECT COST ALLOCATED MODEL
PRODUCTION COSTS MANAGEMENT BY MEANS OF INDIRECT COST ALLOCATED MODEL Berislav Bolfek 1, Jasna Vujčić 2 1 Polytechnic Slavonski Brod, Croatia, berislav.bolfek@vusb.hr 2 High school ''Matija Antun Reljković'',
More informationOptimal Design Of English Auctions With Discrete Bid Levels*
Optial Design Of English Auctions With Discrete Bid Levels* E. David, A. Rogers and N. R. Jennings Electronics and Coputer Science, University of Southapton, Southapton, SO7 BJ, UK. {ed,acr,nrj}@ecs.soton.ac.uk.
More informationASSESSING CREDIT LOSS DISTRIBUTIONS FOR INDIVIDUAL BORROWERS AND CREDIT PORTFOLIOS. BAYESIAN MULTI-PERIOD MODEL VS. BASEL II MODEL.
ASSESSING CREIT LOSS ISTRIBUTIONS FOR INIVIUAL BORROWERS AN CREIT PORTFOLIOS. BAYESIAN ULTI-PERIO OEL VS. BASEL II OEL. Leonid V. Philosophov,. Sc., Professor, oscow Coittee of Bankruptcy Affairs. 33 47
More informationCHAPTER 2: FUTURES MARKETS AND THE USE OF FUTURES FOR HEDGING
CHAPER : FUURES MARKES AND HE USE OF FUURES FOR HEDGING Futures contracts are agreeents to buy or sell an asset in the future for a certain price. Unlike forward contracts, they are usually traded on an
More informationCatastrophe Insurance Products in Markov Jump Diffusion Model
Catastrophe Insurance Products in Markov Jup Diffusion Model (Topic of paper: Risk anageent of an insurance enterprise) in Shih-Kuei Assistant Professor Departent of Finance National University of Kaohsiung
More informationMAT 3788 Lecture 3, Feb
The Tie Value of Money MAT 3788 Lecture 3, Feb 010 The Tie Value of Money and Interest Rates Prof. Boyan Kostadinov, City Tech of CUNY Everyone is failiar with the saying "tie is oney" and in finance there
More informationA Description of Swedish Producer and Import Price Indices PPI, EXPI and IMPI
STATSTCS SWEDE Rev. 2010-12-20 1(10) A Description of Swedish roducer and port rice ndices, EX and M The rice indices in roducer and port stages () ai to show the average change in prices in producer and
More informationResearch on the Management Strategy from the Perspective of Profit and Loss Balance
ISSN: 2278-3369 International Journal of Advances in Manageent and Econoics Available online at: www.anageentjournal.info RESEARCH ARTICLE Research on the Manageent Strategy fro the Perspective of Profit
More informationCapital Asset Pricing Model: The Criticisms and the Status Quo
Journal of Applied Sciences Research, 7(1): 33-41, 2011 ISSN 1819-544X This is a refereed journal and all articles are professionally screened and reviewed 33 ORIGINAL ARTICLES Capital Asset Pricing Model:
More informationDepartment of Econometrics and Business Statistics
ISSN 440-77X Australia Departent of Econoetrics and Business Statistics http://www.buseco.onash.edu.au/depts/ebs/pubs/wpapers/ Applications of Inforation Measures to Assess Convergence in the Central Liit
More informationExtremeVaR of South East Asian Stock Indices with Extreme Distribution- Based Efficiency
Available online at www.sciencedirect.co ScienceDirect Procedia Econoics and Finance 5 ( 2013 ) 120 124 International Conference on Applied Econoics (ICOAE) 2013 EtreeVaR of South East Asian Stock Indices
More informationWho Gains and Who Loses from the 2011 Debit Card Interchange Fee Reform?
No. 12-6 Who Gains and Who Loses fro the 2011 Debit Card Interchange Fee Refor? Abstract: Oz Shy In October 2011, new rules governing debit card interchange fees becae effective in the United States. These
More informationA NUMERICAL EXAMPLE FOR PORTFOLIO OPTIMIZATION. In 2003, I collected data on 20 stocks, which are listed below: Berkshire-Hathaway B. Citigroup, Inc.
A NUMERICAL EXAMPLE FOR PORTFOLIO OPTIMIZATION In 3, I collected data on stocks, which are listed below: Sybol ADBE AMZN BA BRKB C CAT CSCO DD FDX GE GLW GM INTC JNJ KO MO MSFT RTN SBC Nae Adobe Systes
More informationIntroduction to Risk, Return and the Opportunity Cost of Capital
Introduction to Risk, Return and the Opportunity Cost of Capital Alexander Krüger, 008-09-30 Definitions and Forulas Investent risk There are three basic questions arising when we start thinking about
More informationAN ANALYSIS OF EQUITY IN INSURANCE. THE MATHEMATICAL APPROACH OF RISK OF RUIN FOR INSURERS
Iulian Mircea AN ANALYSIS OF EQUITY IN INSURANCE. THE MATHEMATICAL APPROACH OF RISK OF RUIN FOR INSURERS A.S.E. Bucure ti, CSIE, Str.Mihail Moxa nr. 5-7, irceaiulian9@yahoo.co, Tel.074.0.0.38 Paul T n
More informationThe Social Accounting Matrix (SAM)
Università degli Studi di Roa "Tor Vergata The Social Accounting Matrix (SAM) Methodology and Web site Federica Alfani 17 Maggio 2009 The Social Accounting Matrix (SAM) Iportant aspects related to this
More informationSee Market liquidity: Research Findings and Selected Policy Implications in BIS (1999) for the various dimensions of liquidity.
Estiating liquidity preia in the Spanish Governent securities arket 1 Francisco Alonso, Roberto Blanco, Ana del Río, Alicia Sanchís, Banco de España Abstract This paper investigates the presence of liquidity
More informationOPTIMIZATION APPROACHES IN RISK MANAGEMENT: APPLICATIONS IN FINANCE AND AGRICULTURE
OPTIMIZATION APPROACHES IN RISK MANAGEMENT: APPLICATIONS IN FINANCE AND AGRICULTURE By CHUNG-JUI WANG A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
More informationStaff Memo N O 2005/11. Documentation of the method used by Norges Bank for estimating implied forward interest rates.
N O 005/ Oslo Noveber 4, 005 Staff Meo Departent for Market Operations and Analysis Docuentation of the ethod used by Norges Bank for estiating iplied forward interest rates by Gaute Myklebust Publications
More informationProject selection by using AHP and Bernardo Techniques
International Journal of Huanities and Applied Sciences (IJHAS) Vol. 5, No., 06 ISSN 77 4386 Project selection by using AHP and Bernardo Techniques Aza Keshavarz Haddadha, Ali Naazian, Siaak Haji Yakhchali
More informationHandelsbanken Debt Security Index Base Methodology. Version September 2017
Handelsbanken Debt Security Index Base ethodology Version 1.0 22 Septeber 2017 Contents 1 Introduction... 3 2 Description... 3 3 General Ters... 3 4 Iportant Inforation... 4 5 Definitions... 5 5.1 iscellaneous...
More informationWhy Do Large Investors Disclose Their Information?
Why Do Large Investors Disclose Their Inforation? Ying Liu Noveber 7, 2017 Abstract Large investors often advertise private inforation at private talks or in the edia. To analyse the incentives for inforation
More informationResearch on Entrepreneur Environment Management Evaluation Method Derived from Advantage Structure
Research Journal of Applied Sciences, Engineering and Technology 6(1): 160-164, 2013 ISSN: 2040-7459; e-issn: 2040-7467 Maxwell Scientific Organization, 2013 Subitted: Noveber 08, 2012 Accepted: Deceber
More informationCONDITIONAL MEAN DOMINANCE: TESTING FOR SUFFICIENCY OF ANOMALIES
CONDITIONAL MEAN DOMINANCE: TESTING FOR SUFFICIENCY OF ANOMALIES K. Victor Chow and Ou Hu* ABSTRACT Extensive epirical literature of anoalies suggests that an asset reallocation by buying a subset of the
More informationAIM V.I. Small Cap Equity Fund
AIM V.I. Sall Cap Equity Fund PROSPECTUS May 1, 2009 Series I shares Shares of the fund are currently offered only to insurance copany separate accounts funding variable annuity contracts and variable
More informationMonte Carlo Methods. Monte Carlo methods
ρ θ σ µ Monte Carlo Methos What is a Monte Carlo Metho? Rano walks The Metropolis rule iportance sapling Near neighbor sapling Sapling prior an posterior probability Exaple: gravity inversion The ovie
More informationVariance Swaps and Non-Constant Vega
Variance Swaps and Non-Constant Vega David E. Kuenzi Head of Risk anageent and Quantitative Research Glenwood Capital Investents, LLC 3 N. Wacker Drive, Suite 8 Chicago, IL 666 dkuenzi@glenwood.co Phone
More informationEvaluation on the Growth of Listed SMEs Based on Improved Principal Component Projection Method
Proceedings of the 7th International Conference on Innovation & Manageent 519 Evaluation on the Growth of Listed SMEs Based on Iproved Principal Coponent Projection Method Li Li, Ci Jinfeng Shenzhen Graduate
More informationLocal Volatility Models. Copyright Changwei Xiong June last update: October 28, 2017
Local Volatility Models Copyright Changwei Xiong 016 June 016 last update: October 8, 017 TABLE OF CONTENTS Table of Contents...1 1. ologorov Forward and Backward Equations... 1.1. ologorov Forward Equation...
More informationFinancialTimeSeriesRecentTrendsinEconometrics
Global Journal of Manageent and Business Research Finance Volue 13 Issue 5 Version 1.0 Year 013 Type: Double Blind Peer Reviewed International Research Journal Publisher: Global Journals Inc. (USA) Online
More informationUnisex-Calculation and Secondary Premium Differentiation in Private Health Insurance. Oliver Riedel
Unisex-Calculation and Secondary Preiu Differentiation in Private Health Insurance Oliver Riedel University of Giessen Risk Manageent & Insurance Licher Strasse 74, D - 35394 Giessen, Gerany Eail: oliver.t.riedel@wirtschaft.uni-giessen.de
More informationReinsurance and securitization: Application to life risk management. Pauline Barrieu Henri Loubergé
securitization: Application to life risk anageent Pauline Barrieu Henri Loubergé 2 Background Traditional reinsurance theory (Borch, 1960, 1962): a reinsurance pool is fored to share undiversifiable risk;
More informationThe Least-Squares Method for American Option Pricing
U.U.D.M. Proect Report 29:6 The Least-Squares Method for Aerican Option Pricing Xueun Huang and Xuewen Huang Exaensarbete i ateatik, 3 hp + 5 hp Handledare och exainator: Macie Kliek Septeber 29 Departent
More informationLECTURE 4: MIXED STRATEGIES (CONT D), BIMATRIX GAMES
LECTURE 4: MIXED STRATEGIES (CONT D), BIMATRIX GAMES Mixed Strategies in Matrix Gaes (revision) 2 ixed strategy: the player decides about the probabilities of the alternative strategies (su of the probabilities
More informationRealized Variance and IID Market Microstructure Noise
Realized Variance and IID Market Microstructure Noise Peter R. Hansen a, Asger Lunde b a Brown University, Departent of Econoics, Box B,Providence, RI 02912, USA b Aarhus School of Business, Departent
More informationThe Institute of Chartered Accountants of Sri Lanka
The Institute of Chartered Accountants of Sri Lanka Executive Diploa in Business and Accounting Financial Matheatics Financial Matheatics deals with probles of investing Money, or Capital. If the investor
More informationResearch Article Analysis on the Impact of the Fluctuation of the International Gold Prices on the Chinese Gold Stocks
Discrete Dynaics in Nature and Society, Article ID 308626, 6 pages http://dx.doi.org/10.1155/2014/308626 Research Article Analysis on the Ipact of the Fluctuation of the International Gold Prices on the
More informationNontradables and relative price levels across areas within Japan Hidehiro Ikeno Surugadai University
Nontradables and relative price levels across areas within Japan Hidehiro Ieno Surugadai University 1. Introduction This paper exaines epirically the iportance of tradables and nontradables in deterining
More informationAn Application of Extreme Value Theory for Measuring Financial Risk in the Uruguayan Pension Fund 1
An Application of Extreme Value Theory for Measuring Financial Risk in the Uruguayan Pension Fund 1 Guillermo Magnou 23 January 2016 Abstract Traditional methods for financial risk measures adopts normal
More informationStochastic Analysis of Life Insurance Surplus
Stochastic Analysis of Life Insurance Surplus Natalia Lysenko, Gary Parker Abstract The behaviour of insurance surplus over tie for a portfolio of hoogeneous life policies in an environent of stochastic
More informationOn the population median estimation using quartile double ranked set sampling
On te population edian estiation using quartile double ranked set sapling Aer Ibrai Al-Oari Faculty of Science Departent of Mateatics Al al-bayt University Mafraq Jordan aloari_aer@yaoo.co Loai M. Al-Zubi
More informationRecursive Inspection Games
Recursive Inspection Gaes Bernhard von Stengel February 7, 2016 arxiv:1412.0129v2 [cs.gt] 7 Feb 2016 Abstract We consider a sequential inspection gae where an inspector uses a liited nuber of inspections
More informationA Pricing Model for Milk Based on Cost of Production
Tropical Agricultural Research Vol: 9: 3-334 (007) A Pricing Model for Mil Based on Cost of Production V Saravanauar and DK Jain Agribusiness Developent Division Tail Nadu Agricultural University Coibatore
More informationProduction, Process Investment and the Survival of Debt Financed Startup Firms
Babson College Digital Knowledge at Babson Babson Faculty Research Fund Working Papers Babson Faculty Research Fund 00 Production, Process Investent and the Survival of Debt Financed Startup Firs S. Sinan
More informationHistorical Yield Curve Scenarios Generation without Resorting to Variance Reduction Techniques
Working Paper Series National Centre of Copetence in Research Financial Valuation and Risk Manageent Working Paper No. 136 Historical Yield Curve Scenarios Generation without Resorting to Variance Reduction
More informationCompensation Report. Fresenius Medical Care AG & Co. KGaA
Copensation Report Fresenius Medical Care AG & Co. KGaA Copensation Report The copensation report of FMC-AG & Co. KGaA suarizes the ain eleents of the copensation syste for the ebers of the Manageent Board
More informationSection on Survey Research Methods
Using the Statistics of Incoe Division s Saple Data to Reduce Measureent and Processing Error in Sall Area Estiates Produced fro Adinistrative Tax Records Kiberly Henry, Partha Lahiri, and Robin Fisher
More informationQED. Queen s Economics Department Working Paper No. 1088
QED Queen s Econoics Departent Working Paper No. 1088 Regulation and Taxation of Casinos under State-Monopoly, Private Monopoly and Casino Association Regies Hasret Benar Eastern Mediterranean University
More informationExpert Advisor (EA) Evaluation System Using Web-based ELECTRE Method in Foreign Exchange (Forex) Market
The 2 nd International Conference on Energy, Environent and Inforation Syste (ICENIS 2017) 15 th 16 th August 2017, Universitas Diponegoro, Searang, Indonesia Expert Advisor (EA) Evaluation Syste Using
More informationImprecise Probabilities in Non-cooperative Games
7th International Syposiu on Iprecise Probability: Theories and Applications, Innsbruck, Austria, 2011 Iprecise Probabilities in Non-cooperative Gaes Robert Nau Fuqua School of Business Duke University
More informationExtreme Risk Analysis July 2009
Extree Risk Analysis Lisa R. Goldberg Michael Y. Hayes Jose Menchero Indrajit Mitra Quantitative risk anageent allows for qualitative notions such as optiality and expected returns to be put on a quantitative
More informationNBER WORKING PAPER SERIES WEAK AND SEMI-STRONG FORM STOCK RETURN PREDICTABILITY, REVISITED. Wayne E. Ferson Andrea Heuson Tie Su
NBER WORKING PAPER SERIES WEAK AND SEMI-STRONG FORM STOCK RETURN PREDICTABILITY, REVISITED Wayne E. Ferson Andrea Heuson Tie Su Working Paper 10689 http://www.nber.org/papers/w10689 NATIONAL BUREAU OF
More informationUNCOVERED INTEREST PARITY IN CENTRAL AND EASTERN EUROPE: CONVERGENCE AND THE GLOBAL FINANCIAL CRISIS 1
UNCOVERED INTEREST PARITY IN CENTRAL AND EASTERN EUROPE: CONVERGENCE AND THE GLOBAL FINANCIAL CRISIS 1 Abstract Fabio Filipozzi 2, Karsten Staehr Tallinn University of Technology, Bank of Estonia This
More informationm-string Prediction
Figure 1. An =3 strategy. -string Prediction 000 0 001 1 010 1 011 0 100 0 101 1 110 0 111 1 10 Figure 2: N=101 s=1 9 8 7 6 σ 5 4 3 2 1 0 0 2 4 6 8 10 12 14 16 42 Figure 3: N=101 s=2 15 10 σ 5 0 0 2 4
More informationBERMUDA NATIONAL PENSION SCHEME (GENERAL) REGULATIONS 1999 BR 82 / 1999
QUO FA T A F U E R N T BERMUDA NATIONAL PENSION SCHEME (GENERAL) REGULATIONS 1999 BR 82 / 1999 TABLE OF CONTENTS 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Citation Interpretation PART 1 PRELIMINARY PART II REGISTRATION
More informationEconomic Growth, Inflation and Wage Growth: Experience from a Developing Country
www.sciedu.ca/br Business and Manageent Research Vol., No. ; 0 Econoic Growth, Inflation and Wage Growth: Experience fro a Developing Countr Shahra Fattahi (Corresponding author) Departent of Econoics
More informationState Trading Enterprises as Non-Tariff Measures: Theory, Evidence and Future Research Directions
State Trading Enterprises as Non-Tariff Measures: Theory, Evidence and Future Research Directions Steve McCorriston (University of Exeter, UK) (s.ccorriston@ex.ac.uk) Donald MacLaren (university of Melbourne,
More information\Notes" Yuri Y. Boykov. 4 August Analytic approximation of. In this chapter we apply the method of lines to approximate values of several
\Notes" Yuri Y. Boyov 4 August 1996 Part II Analytic approxiation of soe exotic options 1 Introduction In this chapter we apply the ethod of lines to approxiate values of several options of both European
More informationWeak and Semi-strong Form Stock Return Predictability Revisited
Weak and Sei-strong For Stock Return Predictability Revisited WAYNE E. FERSON ANDREA HEUSON TIE SU Boston College 140 Coonwealth Avenue, Chestnut Hill, MA. 02467 University of Miai 5250 University Drive,
More informationAn Analytical Solution to Reasonable Royalty Rate Calculations a
-0- An Analytical Solution to Reasonable Royalty Rate Calculations a Willia Choi b Roy Weinstein c July 000 Abstract The courts are increasingly encouraging use of ore rigorous, scientific approaches to
More informationPerformance Analysis of Online Anticipatory Algorithms for Large Multistage Stochastic Integer Programs
Perforance Analysis of Online Anticipatory Algoriths for Large Multistage Stochastic Integer Progras Luc Mercier and Pascal Van Hentenryck Brown University {ercier,pvh}@cs.brown.edu Abstract Despite significant
More informationTime Varying International Market Integration
International Journal of conoics and Finance; Vol. 5, No. 11; 013 ISSN 1916-971X-ISSN 1916-978 Published by Canadian Center of Science and ducation Tie Varying International Market Integration Dhouha Hadidane
More informationOptimal resource allocation among transit agencies for fleet management
Optial resource allocation aong transit agencies for fleet anageent To V Mathew a, Snehaay Khasnabis b, Sabyasachee Mishra b a Departent of Civil Engineering, Indian Institute of Technology Bobay, Powai,
More informationConstruction Methods.. Ch.-2- Factors Affecting the Selection of Construction Equipment
Construction Methods.. Ch.-2- Factors Affecting the Selection of Construction Equipent Chapter 2 Factors Affecting the Selection of Construction Equipent 2. Factors Affecting the Selection of Construction
More informationImplementation of MADM Methods in Solving Group Decision Support System on Gene Mutations Detection Simulation
Ipleentation of MADM Methods in Solving Group Decision Support Syste on Gene Mutations Detection Siulation Eratita *1, Sri Hartati *2, Retantyo Wardoyo *2, Agus Harjoko *2 *1 Departent of Inforation Syste,
More informationA Complete Example of an Optimal. Two-Bracket Income Tax
A Coplete Exaple of an Optial Two-Bracket Incoe Tax Jean-François Wen Departent of Econoics University of Calgary March 6, 2014 Abstract I provide a siple odel that is solved analytically to yield tidy
More informationNBER WORKING PAPER SERIES THE LEVERAGE EFFECT PUZZLE: DISENTANGLING SOURCES OF BIAS AT HIGH FREQUENCY. Yacine Ait-Sahalia Jianqing Fan Yingying Li
NBER WORKING PAPER SERIES THE LEVERAGE EFFECT PUZZLE: DISENTANGLING SOURCES OF BIAS AT HIGH FREQUENCY Yacine Ait-Sahalia Jianqing Fan Yingying Li Working Paper 17592 http://www.nber.org/papers/w17592 NATIONAL
More informationThe New Keynesian Phillips Curve for Austria An Extension for the Open Economy
The New Keynesian Phillips Curve for Austria An Extension for the Open Econoy Following the epirical breakdown of the traditional Phillips curve relationship, the baseline New Keynesian Phillips Curve
More informationA NOJE;o.n INTEREST RATE RISK, SYSTEMATIC RISK and the PLANNING. Researchmeraorandum Sept. '85
ET 05548 1985 023 SERIE RE5ERR[HE0RHHDH A NOJE;o.n INTEREST RATE RISK, SYSTEMATIC RISK and the PLANNING HORIZON Leon J. de Man Researcheraorandu 1985-23 Sept. '85 VRIJE UNIVERSITEIT Faculteit der Econoische
More informationTotal PS TG. Budgeted production levels can be calculated as follows:
U. ;' cn '.:. \.' >>.:---"--^ '-.'" * i--.'. * ::-;.v>"--:'i.-^ -7 -..=../.-' "-. " '.:.' Ill all it.;? s Solution Total PS TG Sales units 6,000 5,000 1,000 Sales value $605,000 $475,000 $130,000 Workings
More informationMADM Methods in Solving Group Decision Support System on Gene Mutations Detection Simulation
MADM Methods in Solving Group Decision Support Syste on Gene Mutations Detection Siulation Eratita *1, Sri Hartati *2, Retantyo Wardoyo *2, Agus Harjoko *2 *1 Departent of Inforation Syste, Coputer Science
More informationIs FDI Indeed Tariff-Jumping? Firm-Level Evidence
Is FDI Indeed Tariff-Juping? Fir-Level Evidence Ayça Tekin-Koru March 0, 004 Abstract This paper attepts to shed light on greenfield FDI and cross-border M&A as distinct FDI odes of entry. The paper first
More informationQED. Queen s Economics Department Working Paper No Hasret Benar Department of Economics, Eastern Mediterranean University
QED Queen s Econoics Departent Working Paper No. 1056 Regulation and Taxation of Casinos under State-Monopoly, Private Monopoly and Casino Association Regies Hasret Benar Departent of Econoics, Eastern
More informationCorrective Taxation versus Liability
Aerican Econoic Review: Papers & Proceedings 2011, 101:3, 273 276 http://www.aeaweb.org/articles.php?doi=10.1257/aer.101.3.273 Law and Econoics Corrective Taxation versus Liability By Steven Shavell* Since
More informationPuerto Rico, US, Dec 2013: 5-year sentence for pricefixing
Dynaic oligopoly theory Collusion price coordination Illegal in ost countries - Explicit collusion not feasible - Legal exeptions Recent EU cases - Banking approx..7 billion Euros in fines (03) - Cathodic
More informationPing Cao School of Economic and Management, Shanghai Institute of Technology, Shanghai, China
doi:10.21311/001.39.7.27 Payent Scheduling Probles of Software Projects fro a Bilateral Perspective Ping Cao School of Econoic and Manageent, Shanghai Institute of Technology, Shanghai, China Jian Zhang*
More informationOperational Risk Aggregation
Operational Risk Aggregation Professor Carol Alexander Chair of Risk Management and Director of Research, ISMA Centre, University of Reading, UK. Loss model approaches are currently a focus of operational
More informationDependence of default probability and recovery rate in structural credit risk models: Case of Greek banks
Dependence of default probability and recovery rate in structural credit risk odels: Case of Greek banks Abdelkader Derbali, Laia Jael To cite this version: Abdelkader Derbali, Laia Jael. Dependence of
More informationARTICLE IN PRESS. Pricing in debit and credit card schemes. Julian Wright* 1. Introduction
ARTICLE IN PRE Econoics Letters x (200) xxx xxx www.elsevier.co/ locate/ econbase Pricing in debit and credit card schees Julian Wright* Departent of Econoics, University of Auckland, Private ag 92019,
More informationSome estimates for income elasticities of leisure activities in the United States
MPRA Munich Personal RePEc Archive Soe estiates for incoe elasticities of leisure activities in the United States Jorge González Chapela Centro Universitario de la Defensa de Zaragoza 14. July 2014 Online
More informationSpeculation in commodity futures markets: A simple equilibrium model
Speculation in coodity futures arkets: A siple equilibriu odel Bertrand Villeneuve, Delphine Lautier, Ivar Ekeland To cite this version: Bertrand Villeneuve, Delphine Lautier, Ivar Ekeland. Speculation
More informationCREDIT AND TRAINING PROVISION TO THE POOR BY VERTICALLY CONNECTED NGO S AND COMMERCIAL BANKS
CREDIT AND TRAINING PROVISION TO THE POOR BY VERTICALLY CONNECTED NGO S AND COMMERCIAL BANKS Gherardo Gino Giuseppe Girardi Econoics, Finance and International Business London Metropolitan University g.girardi@londoneac.uk
More informationCombining Neural Network and Firefly Algorithm to Predict Stock Price in Tehran exchange
Cobining Neural Network and Firefly Algorith to Predict Stock Price in Tehran exchange Aliabdollahi Departent of Accounting Persian Gulf International Branch Islaic Azad Univercity,khorrashahr, Iran Saharotaedi
More informationCapital reserve planning:
C O - O P E R A T I V E H O U S I N G F E D E R A T I O N O F C A N A D A Capital reserve planning: A guide for federal-progra co-ops Getting our house in order P A R T O F T H E 2 0 2 0 V I S I O N T
More informationEVOLVING PARAMETERS OF LOGIT MODEL USING GENETIC ALGORITHMS
EVOLVING PARAMETERS OF LOGIT MODEL USING GENETIC ALGORITHMS Ming Zhong, University of New Brunswick Pawan Lingras and Will Blades, Saint Mary s University John Douglas Hunt, University of Calgary Introduction
More informationNon-Linear Exchange Rate Pass-Through in Emerging Markets
WP/16/1 Non-Linear Exchange Rate Pass-Through in Eerging Markets by Francesca G. Caselli and Agustin Roitan IMF Working Papers describe research in progress by the author(s) and are published to elicit
More informationNon-Linear Programming Approach for Optimization of Construction Project s Control System
Non-Linear Prograing Approach for Optiization of Construction Project s Control Syste Yusrizal Lubis 1 and Zuhri 1,2 Sekolah Tinggi Teknik Harapan Medan Indonesia, Sekolah Tinggi Ilu Manajeen Suka Medan
More informationRisk Sharing, Risk Shifting and the Role of Convertible Debt
Risk Sharing, Risk Shifting and the Role of Convertible Debt Saltuk Ozerturk Departent of Econoics, Southern Methodist University Abstract This paper considers a financial contracting proble between a
More informationARTICLE IN PRESS. Journal of Mathematical Economics xxx (2008) xxx xxx. Contents lists available at ScienceDirect. Journal of Mathematical Economics
Journal of Matheatical Econoics xxx (28) xxx xxx Contents lists available at ScienceDirect Journal of Matheatical Econoics journal hoepage: www.elsevier.co/locate/jateco 1 1 2 2 3 4 5 6 7 8 9 1 11 12 13
More informationProvided in Cooperation with: Center for Financial Studies (CFS), Goethe University Frankfurt
econstor www.econstor.eu Der Open-Access-Publikationsserver der ZBW Leibniz-Inforationszentru Wirtschaft The Open Access Publication Server of the ZBW Leibniz Inforation Centre for Econoics Schidt, Daniel
More informationStrategic Second Sourcing by Multinationals. Jay Pil Choi and Carl Davidson Michigan State University March 2002
trategic econd ourcing by Multinationals Jay Pil Choi and Carl Davidson Michigan tate University March 2002 Abstract: Multinationals often serve foreign arkets by producing doestically and exporting as
More information- Chapter 2 - LINKING CGE AND MICROSIMULATION MODELS: A COMPARISON OF DIFFERENT APPROACHES
- Chapter 2 - LINKING CGE AND MICROSIMULATION MODELS: A COMPARISON OF DIFFERENT APPROACHES 72 . INTRODUCTION In the literature that studies incoe ineuality and poverty, we can observe a recent developent
More information