Espinosa et al.

Health Economics Review (2023) 13:15 Health Economics Review

https://doi.org/10.1186/s13561-022-00416-5

RESEARCH Open Access

Statistical actuarial estimation

of the Capitation Payment Unit from copula
functions and deep learning: historical
comparability analysis for the Colombian health
system, 2015–2021
Oscar Espinosa1 , Valeria Bejarano1* , Jeferson Ramos1   and Boris Martínez2   

Abstract
The Capitation Payment Unit (CPU) financing mechanism constitutes more than 70% of health spending in Colombia,
with a budget allocation of close to 60 trillion Colombian pesos for the year 2022 (approximately 15.7 billion US dol-
lars). This article estimates actuarially, using modern techniques, the CPU for the contributory regime of the General
System of Social Security in Health in Colombia, and compares it with what is estimated by the Ministry of Health and
Social Protection. Using freely available information systems, by means of statistical copulas functions and artificial
neural networks, pure risk premiums are calculated between 2015 and 2021. The study concludes that the weights by
risk category are systematically different, showing historical pure premiums surpluses in the group of 0–1 years and
deficits (for the regions normal and cities) in the groups over 54 years of age.
Keywords Pure risk premium, Health system, Copulas, Artificial neural networks, Actuarial science
JEL C45, C51, G22, I13

Introduction currently calculated by the Ministry of Health and Social

The General System of Social Security in Health (Sis- Protection (MHSP), has, since 2006, been computed
tema General de Seguridad Social en Salud, SGSSS for based on three variables (risk adjusters): age, sex and
its acronym in Spanish) of Colombia has different financ- region [25].
ing mechanisms for its operation; the most important in The purpose of the CPU is to finance a set of health
terms of monetary magnitudes is the so-called Capitation technologies (drugs, procedures, supplies, medical
Payment Unit (CPU). This ‘health insurance premium’, devices, etc.), known as the Health Benefits Plan (HBP-
CPU), which configures a collective protection mecha-
nism for the right to health under a mandatory insurance
*Correspondence: scheme [9].
Valeria Bejarano
For example, for the year 2020, the SGSSS Resources
[email protected]
1
Economic Models and Quantitative Methods Research Group, Centro Administrator (ADRES for its acronym in Spanish), the
de Investigaciones para el Desarrollo, Universidad Nacional de Colombia, entity in charge of making the recognition and payment
Bogotá, D.C., Colombia
2 of the CPU to health insurers (called Entities Adminis-
Department of Mathematics, Universidad Nacional de Colombia,
Bogotá, D.C., Colombia trators of Health Benefit Plans, EAHBP), made transfers
of about 48.5 trillion Colombian pesos (COP) for the

© The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which
permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this
licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/. The Creative Commons Public Domain Dedication waiver (http://​creat​iveco​
mmons.​org/​publi​cdoma​in/​zero/1.​0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Espinosa et al. Health Economics Review (2023) 13:15 Page 2 of 20

contributory (CR)1 and subsidized (SR)2 regimes, distrib- 55–59 years, 60–64 years, 65–69 years, 70–74 years, and
uted in similar proportions [1]. 75 years or more.
In a context of budgetary restrictions – common to all This paper is structured as follows. The first section
countries, of any income level – and in the face of an evi- presents the historical context for the CPU and its pric-
dent growing demand for more and better health tech- ing in the SGSSS. The second section offers a descriptive
nologies for the inhabitants, the financial sustainability analysis of the data of interest and the new methodo-
of the SGSSS must be ensured, maximizing as far as pos- logical proposal for estimating the statistical-actuarial
sible the results in terms of the health of the population pricing models. The third section presents the most rel-
of the entire national territory. The pressures of health evant results and findings on the actuarial variables of
spending derived from the extensions of the HBP-CPU frequency, severity and pure risk premiums. Finally, the
are a constant challenge for health systems, therefore, the fourth section contains the final considerations of the
constant study of the sufficiency of the cost of health risk research, its main limitations and some proposals for
management should be a priority evaluation issue for the future research on the subject.
care of state finances.
In this scenario, the analysis of the CPU’s pricing Historical context of the CR‑CPU and its pricing
becomes relevant. In this regard, the specialized litera- The social bodies responsible for establishing the values
ture has investigated alternatives for risk adjustment in of the CPU have been in historical order: the National
SGSSS health spending [4, 21, 43, 44]. However, no stud- Council for Social Security in Health, the Health Regula-
ies have been found that develop a particular method for tion Commission, and (currently) the Directorate for the
calculating the CPU of the risk groups defined by the leg- Regulation of Health Insurance Benefits, Costs and Rates
islation. The only antecedents are the official documents of the MHSP. Since 2010, unlike previous years, the suffi-
of the MHSP and the investigation by Basto et al. [3], ciency studies use a clear actuarial concept, based on the
which focuses exclusively on SR. fundamental insurance equation, assuming that the CPU
Because of this, the present research aims to estimate can be understood as the division between the expected
actuarially the pure risk premiums for CR3 by means of value of health costs and the population exposed to
copulas functions and deep learning approximations, and health risk [53].
to compare the estimated monetary values with those From the statistical-actuarial approach, pricing meth-
defined by the resolutions, for the years 2015 to 2021. ods are used to build premiums that cover the losses of
[27–30, 32, 34, 36]. This will allow reviewing and con- the insured’s subscribed risks, that is, that are sufficient,
trasting the budget allocations that have been made over with a high degree of confidence [6, 12]. To estimate
time based on real-world evidence and taking note of the CPU rate, the MHSP has used the method called
possible improvements in the computation of the finan- the expected loss ratio, which is based on the quotient
cial calculation of health risk management in Colombia. between the calculated loss ratio and the permissible
From 2015 to date, the MHSP has estimated the pure loss ratio of the EAHBP (which according to Law 1438 of
risk premiums for 56 groups that categorize the popula- 2011 is of the order of 0.9 for CR). The result indicates
tion affiliated with the health system. For this reason, what is the necessary increase of the CPU to guarantee
the analysis period starts from that year and the esti- the financial sufficiency of the SGSSS [35].
mates are made using the same groups. The 56 groups In this context, the MHSP projects costs, income and
consist of the combinations of the categories of the vari- those exposed to risk. For the first variable, it applies dif-
ables: i) region: normal, remote, cities and special and ii) ferent trend adjustment factors to emulate future condi-
age/sex group: less than 1 year, 1–4 years, 5–14 years, tions: increases in the price level, frequency of claims,
15- 18 years (men), 15–18 years (women), 19–44 years claims that are incurred but not reported (IBNR), HBP-
(men), 19–44 years (women), 45–49 years, 50–54 years, CPU update, among others. For the second variable, it
projects the possible items that make up the income of
the EAHBP of the CR: income from CPU, copayments,
moderating fees, recoveries from the Occupational Risk
Administrators, income from registration and affilia-
1
Of which people who have the ability to pay and contribute jointly and sev- tion fees, income from the High-Cost Account, income
erally to the SGSSS are part (their respective beneficiaries are also included).
2 Agreement 026 of 2012, as well as income from health
People who cannot pay their affiliation to the SGSSS (essentially people
in conditions of vulnerability and poverty) are welcomed here, being subsi- promotion and prevention, among others. For the popu-
dized by the State. lation exposed to risk, the MHSP makes adjustments for
3
Unfortunately for the SR there is no public financial information to cal- missing compensation and for the expected growth in
culate its CPU. Historically, the EAHBPs belonging to this regime have had
significant quality problems in their administrative records [31, 33].
the following year based on the population projections

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Espinosa et al. Health Economics Review (2023) 13:15 Page 3 of 20

of the National Administrative Department of Statistics its acronym in Spanish) was used through the Integrated
(Departamento Administrativo Nacional de Estadística, Social Protection Information System (SISPRO for its
DANE for its acronym in Spanish). acronym in Spanish), which includes all the expenses
The base information for the analysis of the regulatory charged to the HBP-CPU by the EAHBP that exceed the
entity refers to the calendar year immediately prior to the validation meshes of the MHSP. GD can be considered
year of its realization, for example, the sufficiency study a Sufficiency proxy, a confidential database and a funda-
for the year 2019 estimates the increase in the CPU that mental input for the calculation of the CR-CPU from the
will be sufficient during the year 2020 to finance health regulatory entity.
technologies, using real-world data from the year 2018. Both BDUA and GD present disaggregations by sex,
The latter are extracted, among other databases, from municipality code, department, among others, which
the reports on the provision of health services per affili- allows the feasibility of this actuarial calculation, in
ate issued by the EAHBP, the affiliate and compensation accordance with the guidelines and risk adjusters pre-
databases of the CR, the financial statements reported by established in national legislation.
the EAHBP to the entity for inspection, surveillance and Figure 1 shows the frequency of people served and
control (National Health Superintendency) and the tariff the number of people exposed of CR by region and year.
manuals for health technologies financed by the CPU. It can be seen that the frequency is higher in the city
Now, for the case of this study, the conceptual approach and normal regions, and is lower in special and remote
considered to estimate the pure health risk premium regions. The average frequency from 2013 to 2019 was
is the product of frequency and severity, where the first 88.4% in the normal region, followed by cities with 86.7%,
factor corresponds to the ratio between the number of special with 81.0% and remote with 68.9%. The ranges for
distinct people served and those exposed to health risk, each region over the seven years were: remote (58.1%-
4
while the second factor is defined as the ratio between 79%), cities (82.3%-92.4%), special (75.4%-88.9%), and
the total costs of health technologies over the number of remote (85.5%-92.4%).
distinct people served. Formally: The observed frequency associated with the age groups
is shown in Fig A1 (see Appendix). There are no drastic
changes in its evolution over time. On average, the age
Pure health risk premium = Frequency ∗ Severity
⏟⏞⏞⏟⏞⏞⏟ ⏟⏟⏟ groups with the highest frequency were in this order:
Distinct people served
Exposed to health risk
Total costs of health technologies
Distinct people served
1 to 4 years, less than one year, 75 years or older, 70 to
(1) 74 years, 19 to 44 years (women) and 65 to 69 years, these
This classic actuarial approach, unlike the expected loss values are included within the range of 89.5% to 100%. In
ratio, allows the two variables of interest that describe the addition, the age group from 15 to 18 years (men) had the
health risk to be modeled independently and specifically lowest frequency of people attended. On the other hand,
and to project a sufficient CR-CPU. The pure risk pre- the percentage variation of the frequencies between 2013
mium estimated in this way meets the theoretical prop- and 2019 was -0.7% in 15 to 18 years (women), -1.86% in
erties desired in all premiums: additivity, independence, 19–44 years (women) and -2.70% in 19 -44 years (men).
scale invariance, consistency and acceptability [55]. Figure 2 presents the severity (in 2020 prices, COP) and
the number of exposed by region in the CR. From 2013 to
Data and empirical strategy 2019, the remote region presents the greatest severity, on
Data average, 1.34 million COP, followed by cities with 1.1 mil-
For the statistical-actuarial estimation of the CR-CPU, it lion, normal with 0.9 million and special with 0.7 million.
was necessary to have information on: i) those exposed to During this period, severity in the remote region grew
risk (equivalent population), from 2013 to 2020; ii) num- 7.9% in real terms, in cities 11.5%, in normal region 44.9%
ber of distinct people served by the SGSSS, from 2013 and in the special region 43.6%.
to 2019, and iii) severity (average costs) of health care, With regard to severity by age group, Fig A2 (see
from 2013 to 2019. For the first variable, the Database of Appendix) shows that, from 2013 to 2019, it is greater in
Affiliates (Base de Datos Única de Afiliados, BDUA for its groups under one year of age and groups over 60 years
acronym in Spanish) was used, which contains the infor- of age. Severity maintains a stable value over time for all
mation of the fully identified affiliates of the SGSSS who age groups, except for those under one year of age and
are covered by the HBP-CPU; for the second and third, those over 70 years of age, where it decreased until 2015
Demand Management (Gestión de la Demanda, GD for and then increased until 2019. During this time inter-
val, the severity in minors for one year was, on average,
4
1.8 million COP; in the group from 0 to 4 years, 0.7 mil-
Understanding by someone exposed to risk, an individual who was affiliated
with the CR of the SGSSS for a full calendar year.
lion COP; in the groups of men and women from 15 to

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Espinosa et al. Health Economics Review (2023) 13:15 Page 4 of 20

Fig. 1 Frequency of people served by the CR by region, 2013–2019

Fig. 2 CR severity by region, 2013–2019 (2020 prices)

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Espinosa et al. Health Economics Review (2023) 13:15 Page 5 of 20

Fig. 3 Number of people exposed to CR risk, 2013–2019

18 years and 19 to 44 years it was between 0.4 and 0.8 Empirical strategy
million COP; in the ages between 45 and 59 years it was In a first stage, the forecasts of those exposed to the risk
around 1.0 and 1.5 million COP; and in groups over are presented, to later detail the process of computation
60 years of age it ranged from 2.4 million to 3.8 million of the adjustment factors for severity and frequency.
COP. Afterwards, the explanation of the copula functions and
Figure 3 shows that the number of people exposed to the approach taken for the pricing process of the pure
risk in the CR has grown from 2013 to 2019. In 2013 risk premium of the CR is deepened.
there were 19.5 million exposed and in 2019 22.3 million,
which means a growth of 13.9% over the seven years of Forecasts for those at risk
analysis. A deep learning technique called artificial neural net-
Figure 4 shows the distribution of the number of works (ANN) is used, with high predictive power in
exposed according to the region between 2013 and 2019. demographic, financial and health topics [2, 23, 24, 41,
Cities had, on average, 75% of the total exposed, normal 45, 52]. This type of nonlinear nonparametric model is
21.2%, special 3.6% and remote only 0.2%. The propor- considered a self-adaptive, accurate method that requires
tion of those exposed by region was similar throughout very few assumptions. By simulating the operating sys-
the period. tem of a biological neuron, ANNs allow for a flexible
Finally, Fig A3 (see Appendix) represents the participa- approach in terms of corresponding functional forms
tion of the age groups in the number of exposed to CR [52]. Thus, the basic architecture of a three-layer fed-
risk from 2013 to 2019. On average, the participation forward ANN (one input, one hidden, and one output) is
in the total number of exposed of the group from 0 to made up of a set of inputs, weights, activation functions,
4 years is 5.9%, of the group from 5 to 14 years is 13.8%, and outputs. Formally:
of the men and women from 15 to 44 years is 24.4%, rs
( r )
from 45 to 49 years is 18.1% and of the group older than ∑ ∑
2
x=
̂ 𝜑l ⋅ f 𝜑lk xk + 𝜗lk + 𝜗2 , (2)
60 years is 13.2%. The transition towards aging explains l=1 k=1
the greater growth in the participation of older age
groups. The percentage change between 2013 and 2019 where xk (k = 1, … , r) is considered the input set,
in the proportion of those exposed to CR was -12% in the x the output set, f the activation function, 𝜗lk , 𝜑lk y 𝜑j
̂
group from 0 to 4 years, -15.2% from 5 to 14 years, 3.8% (l = 1, … , r) the model parameters and weights, and
in men from 15 to 44 years, 0.4% in women from 15 to rs the neurons in the hidden layers [5]. Then, the ANN
44 years, 1.1% from 45 to 59 years and 15.5% in the group training is based on iteratively adjusting these parame-
over 60 years. ters, so that an error function between the forecast ̂
x and

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Espinosa et al. Health Economics Review (2023) 13:15 Page 6 of 20

Fig. 4 Distribution of those exposed to CR risk by region, 2013–2019

the observation x is minimized. This, from the weighted ii. Inclusion of technologies in the HBP-CPU: it rec-
sum of the outputs of the neurons of the hidden layer. ognizes the new basket of sanitary technologies
This data science technique is used to forecast, based that must be financed in t + 2, considering the
on historical series, those exposed to risk for the 56 cat- actualization/extension of the HBP-CPU that is
egories defined by the CPU and the adjustment factors. made every year.
iii. Comparable: the actual normative is able to finance
sanitary technologies not financed by the CPU but
Adjustment factors for severity and frequency are considered as comparable with some of these;
In the statistical-actuarial process of pricing it is neces- in this case the difference between that technology
sary to express not only the mean cost of attention per and its comparable is recovered.
person but also the amount of people in terms of the tar- iv. Variation in the number of attentions per user: it
get year. Likely MSHP this investigation takes evidence of projects the average number of times that a person
the real world in year t to transform the frequency and receives health technologies.
severity to t + 2, so the economic-financial conditions of v. Inflation: it recognizes the rise in the price levels.
the health system that are expected in the future for the
CR in the country can be represented. For the first three adjustments of severity, the same
The method for constructing these adjustment factors information from the sufficiency studies of the regulatory
developed by Basto et al. [3] is closely followed: the fol- entity is used. For the fourth, using GD, the number of
lowing five factors are employed for the severity: monthly attentions per user is forecast, then averaged for
the months of year t + 2. For the fifth factor, forecasts of
i. Costs incurred but not reported (IBNR): adjust the the Banco de la República (the central bank of Colombia)
monetary amount of attention that the EAHBP did and the Ministry of Finance and Public Credit (MFPC)
not register by the end of the year. are taken.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Espinosa et al. Health Economics Review (2023) 13:15 Page 7 of 20

On the other hand, for frequency, two adjustments are ii)( ∀u1), u2 , v1(, v2 ∈ )[0,1] ( with ) u1 ≤
( u2 , v)1 ≤ v2 then
considered: C u2 , v2 − C u1 , v2 − C u2 , v1 + C u1 , v1 ≥ 0. The
first property shows that the contour region of the copula
i. Effective coverage advance: recognizes the incre- is the consequence of the uniform margin distributions;
ment of the rate between users and exposures, this the second states that C(u, v) is not decreasing in u and v .
proportion has been increasing in the last years. The Sklar theorem (1959) [50] shows that the joint distri-
ii. Changes in the burden of disease: it adjusts the bution H can be expressed in terms of the marginal dis-
appearance of new users attended that had not tributions F and G , and a copula C such that ∀x, y ∈ ℝ:
used the health system, due to the occurrence of ( ) ( ) ( )
H x, y = Pr X ≤ x, Y ≤ y = Pr F −1 (U ) ≤ F −1 (u), G −1 (V ) ≤ G −1 (v)
new health conditions (i.e. new infectious diseases). ( ( ))
= Pr(U ≤ u, V ≤ v) = C(u, v) = C F (x), G y ,
(4)
The first frequency factor is forecast monthly taking
where U = F (X) and V = G(Y ) with U , V ∼ U (0,1),
the information of BDUA and GD, then averaging for the
with F and G as well as their inverse functions monotonic
months of year t + 2. In the case of the second factor, a sim-
increasing. Moreover, if the marginal distribution func-
ilar quantitative operation is made but taking as a proxy the
tions
( are (continuous, then
( there
) exists a unique copula
variable of diagnostics per capita (ICD-10) from GD. ))
C F (x), G y equal to H x, y . The detailed implications
of the different statistical properties can be reviewed in
Pricing with statistical copulas
Nelsen [37].
In the field of actuarial science, copulas have started to
In practice, the most used copula families are Gaussian,
obtain a preponderance at the end of the last century
t-Student, mixed Gaussian and Archimedean. In the last,
and the first decade of the current century, due to their
Gumbel, Clayton and Frank stand out.5 The Gaussian
benefits, in particular the high flexibility of modeling the
and t-Student copulas are derived from their own mul-
joint distribution of a random n-tuple [7, 8, 14, 15]. This
tivariate distributions, for which reason they are called
statistical technique has been applied in several fields
implicit copulas; they also present symmetric depend-
of investigation related to the payment of claims, pric-
ence but are null or low in the tails [38]. On the other
ing, active valorization and, with less relevance, stockpile
hand, the Archimedean copulas are constructed from a
computation, highlighting the opportunity to model the
function 𝜑𝜃 ∶ [0,1] → [0, ∞] that is continuous, mono-
asymmetric dependence in the tails [11, 19, 20, 46, 47,
tone decreasing and convex such that 𝜑𝜃 (1) = 0, where
51]. More recently, copulas have been applied in collec-
𝜑𝜃6 is referred to as the generator function. Additionally,
tive risk models and deductible price-fixing, furthermore,
they describe a great variety of dependence structures,
improvements in the computational efficiency and how to
in particular, they allow modeling asymmetric relations
provide intuitive interpretations of the dependence struc-
between random variables [22, 37].
ture have been investigated [13, 39, 48]. For the sector of
For the computation of the CR-CPU, defining X as
health insurance, the applications in the scientific indexed
severity (continuous variable) and Y as frequency (dis-
literature have been few [49, 54, 56], and in that way, this
crete variable), it is proposed to model the pure risk pre-
work can also be considered a pioneer in the field.
mium by a copula, in this case, mixed. The dependence
In formal terms, and in a succinct way, a copula is a func-
between both variables, following the method developed
tion that describes the dependence between the marginal
by Parra [40], includes different covariables through gen-
probability distributions of two or more random variables
eralized linear models (GLM) in its marginals, which
and is expressed in terms of a multivariate distribution
means
function. In the bivariate case, let (X, Y ) the random vec-
( ) ( ) ∑
tor( )with marginal distributions F (x) = Pr(X ≤ x) and Xi ∼ F xi |𝜇i , 𝜎 ; ln 𝜇i = 𝛼0 + 𝛼l Region
G y = Pr(Y ( ≤)y), respectively, and the (joint) distribution ∑
l
(5)
function H x, y = Pr(X ≤ x, Y ≤ y) for x, y ∈ ℝ2 where + 𝛼k Age∕sex_group,
F , G, H ∼ U (0,1), the bivariate copula C is a function of k
the uniform random variables u = F (x) and v = G(y) that
are constructed in the following way [18, 38]:

C
( ∶ [0,1](×))
[0,1] →([0,1]
), (3)
F (x), G y ↦ H x, y 5
A wide review of other types of copula can be found in Nelsen [37] and
Latorre [22].
and satisfies two properties: i) ∀u, v ∈ [0,1] then 6
𝜃 is the parameter of the Arquimedean copula, which is defined in the
C(u, 0) = 0 = C(0, v), C(u, 1) = u and C(1, v) = v ; bivariate case as C(u, v) = 𝜑−1
( )
𝜑𝜃 (u) + 𝜑𝜃 (v) .
𝜃

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Espinosa et al. Health Economics Review (2023) 13:15 Page 8 of 20

Table 1 Evaluation measures of the selected statistical-actuarial models, 2015–2021

CR-CPU actuarial model 2015 2016 2017 2018 2019 2020 2021

Frequency distribution Negative Binomial Poisson Poisson Negative Binomial Negative Binomial Poisson Poisson
Severity distribution Weibull Normal Normal Log-normal Log-normal Log-normal Log-normal
Copula type t-student Gumbel Joe Frank Frank Frank Frank
MSE 2.26E + 10 6.83E + 10 1.82E + 11 2.55E + 09 1.95E + 09 1.36E + 11 2.01E + 10
MAPE 10.7003 11.6370 8.2159 2.9183 2.8590 9.0923 6.5798
RSCE 4.10E-04 4.34E-04 1.97E-04 2.20E-04 5.34E-04 1.96E-04 9.25E-04
xvCIC 2.1730 0.0000 3.94E-08 1.0037 2.5235 2.8837 5.6776
RGOFC 0.9236 0.8566 0.3372 0.6688 0.8357 0.8077 0.8487
Borda rule result 65 75 75 46 58 57 58

( ) ( ) ∑
Yi ∼ G yi |𝜆i ; ln 𝜆i = 𝛽0 + 𝛽l Region In each year the best model is selected according to the
l Borda’s rule, which order and rank the 72 rival models
∑ according to the values of i) mean square error (MSE); ii)
+ 𝛽k Age∕sex_group (6)
k
mean absolute percentage error (MAPE); iii) the square
root of the square differences between the estimated cop-
+ offset(Exposures).
ula and empirical copula (RSCE) described by Novales
Then, the couple is made by the copula and the joint [38], and iv) the cross-validation copula information cri-
density function of X and Y is found, terion (xvCIC) developed by Grønneberg & Hjort [17].
( ) ( ( ) ( )) For each of these criteria, the best model receives 1 point,
H xi , yi = C F xi |𝜇i , 𝜎 , G yi |𝜆i , (7) the second, 2, and so on.
On the other hand, a regularized goodness of fit test
( ) ( ) [ ( ( ) (
h xi , yi |μi , 𝜎, λi = f xi |μi , 𝜎 ∗ D G yi |λi |F xi |μi , 𝜎
)) is applied for copulas (RGOFC) created by Genest et al.
( ( ) ( ))] (8) [16] based on a statistic of the Anderson–Darling type,it
− D G yi − 1|λi |F xi |μi , 𝜎 ,
has a null hypothesis (Ho ) that the copula presents a
where D(v|u) is the conditional copula of v given u good fit. Here a value of one is assigned if at a signifi-
defined as 𝜕u .
𝜕C(u,v)
cance level of 5% the null hypothesis is rejected and
From Eq. (8) the likelihood is found, and supposing zero in the contrary case. Finally, the winning model for
independence between the observations, the param- each year is the one that has the least total points after
eters of interest of the GLM and the copula are jointly summing the points obtained for these five metrics. In
estimated by its maximization with optimization tech- Table 1 are shown the results of the five metrics of the
niques. Once the final parameters are obtained, Monte chosen models for each year in which the CR-CPU is
Carlo techniques are applied to find values for the ran- estimated.
dom variable from samples of the density function. In The values in COP, of the pure premium estimated, can
the present work, 300 samples are simulated (enough be observed graphically in Fig. 5.7 There, clear historical
to guarantee convergence) and the median is taken as patterns are evidenced in relation to the pure premium
a punctual observation, given its robustness features. estimated by MSHP8 for each year. In summary, as the
Likewise, intervals are constructed from the 2.5 and 97.5 first point to stand out, for every region, in every year the
percentiles. pure premium for the group of less than 1 year given by
MSHP is higher than that computed in this work.
Results
72 statistical-actuarial models are estimated by year.
They come from the combination of the three compo-
7
nents, i) severity distributions: Normal, Weibull, Log- The monetary values of the pure risk premium, the proportion of distinct
persons for each reference year and the values of the frequency/severity
normal, Gamma, Inverse Gamma and Inverse Gaussian; adjustment factors for each pricing year are presented in Appendix A (Tables
ii) frequency distributions: Poisson and Negative Bino- A1, A2, A3, A4, A5, A6, A7).
8
mial and iii) copula types: two implicit (normal and The pure premium of the MHSP is found by multiplying the values of the
CR-CPU set out in the resolutions of the entity by the percentage spent by
t-Student) and four Archimedean (Clayton, Gumbel, the health insurance not allocated to utilities and administration, which is
Frank and Joe). 0.90 in the case of the studied regime [26].

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Espinosa et al. Health Economics Review (2023) 13:15 Page 9 of 20

Second, in the regions ‘cities’ and ‘normal’, in ages is noted that the difference in the estimates for this age
15–18 years (women and men), 65–69 years, 70–74 years group is accentuated with the passing of the years in the
and more than 74 years, the pure premium estimated by remote region.
this study is higher than the one computed by MSHP. As a limitation of this study, the approximation here
As a third point to take into account, for the remote developed is only made for CR, since SR information
and special regions, the pure premium of MSHP is of spending on health technologies has always had
higher although only slightly than the one estimated by problems of bad quality and little representation, for
copulas in ages 19–44 years, 45–49 years, 50–54 years. which reasons there is no data available. It is important
55–59 years, 60–64 year, 70–74 years and more than to remember that this regime, for 2020, had approxi-
74 years.9 mately 23.9 million affiliates and the financing mecha-
nism of the CPU reached values near 24.4 trillion COP
Discussion and conclusions [1]. Thence the importance of paying attention to the
The present investigation had the objective to estimate statistical-actuarial estimations with evidence from the
actuarially the CR-CPU in the SGSSS of Colombia, in a real world.10
systematic and strict way, for the period from 2015 to An adequate estimate of future health spending, as
2021, using modern statistical techniques such as cop- well as the application of efficient risk management
ulas and ANNs. Regarding the sufficiency studies of mechanisms (from a comprehensive approach) and
the CPU developed by the regulatory entity, this work health technology assessments, will allow better long-
is differentiated in the following topics: i) to com- term financial sustainability in national public budgets
pute the pure risk premium, severity and frequency for the health of the population [10, 42]. The meth-
are modeled, then copulas are applied with the pur- odological development presented here contributes to
pose of defining the relation of its joint dependence; the international literature in actuarial health sciences,
ii) to forecast the exposures, analytic approximations showing innovative analytical developments that may
of deep learning are used, which show benefits over become applicable in other countries with pluralis-
other demographic forecast methodologies; iii) good- tic health insurance systems. Likewise, this research
ness of fit criteria and capacity of forecast are used to based on the use of real-world evidence demonstrated
select the best estimations and iv) the adjustment fac- the versatility and functionality of statistical copulas
tors of Basto et al. [3] for severity and frequency are (as an inferential modeling technique), which can con-
considered. tribute to informed decision-making in sector financ-
For the period 2015–2021, in all regions, the estimated ing policy.
pure premium is very close to the pure premium defined Finally, it is important to indicate that this quantita-
by the MSHP in the age groups 5–14, 15–18 (men and tive study is supported and sustained from a prospective
women), 19–44 (men and women). Discrepancy is only approach of computing using the historical data about
observed in the 15–18 group in the remote region in the spending on health technologies financed with the
2017 and 2020 and in the cities region in 2016. CR-CPU. Nonetheless, the ideal scenarios for complete
Compared to the authors’ estimates, the MHSP under- effective coverage and integral health services lend-
estimated the CPU in age groups 55 years and older in ing (meaning, a CPU from an opportunity/normative
the remote region for the years 2017, 2018 and 2019, in approach) is not within the reach of the actual investiga-
the cities region for the years 2015 to 2021, in the nor- tion. This last point will require future investigation pro-
mal region for the years 2018, 2020 and 2021. Instead, the jects that treat these problems with specificity and the
premium is overestimated in age groups over 55 years in corresponding scenarios. In addition, the authors con-
the special region for 2016 and 2017. The difference in sider it wise to review in the future the values of the risk
the estimates for this age group for 2020 are mainly in the weights under a Bayesian approach, which could contrib-
remote and normal regions. ute a certain value-added at the time of adjusting the risk
Surpluses are observed in the estimated pure pre- categories, beyond the benefits already explained that
mium of the MHSP in the group of less than1 year for the result from the use of the statistical copulas presented in
entire period in all regions, mainly remote and special. It this work.

9 10
The analysis developed here make reference to the punctual estimations In the year 2021, for the first time in the history of the country, the MHSP
of the pure premium for the different categories, not to its confidence inter- used proper information of the SR in the actuarial estimation of the CPU of
vals. 2022.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Espinosa et al. Health Economics Review (2023) 13:15 Page 10 of 20

Fig. 5 Pure premiums estimated via copulas for the years 2015 to 2021, versus what was calculated by the MSHP

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Espinosa et al. Health Economics Review (2023) 13:15 Page 11 of 20

Appendix
Figs. A1, A2, and A3.

Fig. A1 Frequency of people served by the CR by age/sex group, 2013–2019

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Espinosa et al. Health Economics Review (2023) 13:15 Page 12 of 20

Fig. A2 CR severity by age/sex group, 2013–2019 (2020 prices)

Fig. A3 Distribution of those exposed to CR risk by age/sex group, 2013–2019

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Espinosa et al. Health Economics Review (2023) 13:15 Page 13 of 20

Tables A1, A2, A3, A4, A5, A6, and A7.

Table A1 Proportion of distinct persons for each reference year

Region Age/sex group 2013 2014 2015 2016 2017 2018 2019

Remote Under 1 year 0.6926 0.8258 0.8961 0.9185 0.9138 0.6158 0.7600
Remote 1–4 years 0.9075 0.9309 0.9867 0.9904 1.0000 0.7469 0.9244
Remote 5–14 years 0.6582 0.5887 0.5737 0.6841 0.7292 0.5222 0.7450
Remote 15–18 years, Men 0.5434 0.4374 0.4043 0.5506 0.5773 0.4176 0.5945
Remote 15–18 years, Women 0.6444 0.5554 0.4818 0.6785 0.7145 0.5321 0.7433
Remote 19–44 years, Men 0.5619 0.5076 0.4817 0.5543 0.6370 0.4222 0.5878
Remote 19–44 years, Women 0.8178 0.7424 0.7240 0.8299 0.9041 0.6634 0.8398
Remote 45–49 years 0.7035 0.5942 0.5427 0.6704 0.7120 0.5233 0.6744
Remote 50–54 years 0.7359 0.6058 0.5560 0.7207 0.7617 0.5927 0.7479
Remote 55–59 years 0.7588 0.6778 0.5652 0.7540 0.8203 0.6231 0.7723
Remote 60–64 years 0.7957 0.6919 0.6192 0.7275 0.8478 0.6726 0.7930
Remote 65–69 years 0.7957 0.7137 0.6755 0.8286 0.8950 0.7337 0.8584
Remote 70–74 years 0.8173 0.7486 0.6651 0.8280 0.9029 0.7323 0.8588
Remote 75 years and older 0.8558 0.8000 0.7102 0.8329 0.9168 0.7659 0.9119
Cities Under 1 year 0.9342 1.0000 1.0000 0.9869 1.0000 0.9391 0.9213
Cities 1–4 years 1.0000 1.0000 1.0000 1.0000 0.9915 0.9600 0.9623
Cities 5–14 years 0.8281 0.8730 0.8800 0.8091 0.8297 0.7909 0.8270
Cities 15–18 years, Men 0.7344 0.7751 0.7532 0.6871 0.7081 0.6673 0.6971
Cities 15–18 years, Women 0.8959 0.9598 0.9342 0.8385 0.8677 0.8019 0.8267
Cities 19–44 years, Men 0.7834 0.7991 0.7946 0.7287 0.7200 0.7069 0.7161
Cities 19–44 years, Women 0.9950 1.0000 1.0000 0.9197 0.9111 0.8935 0.9041
Cities 45–49 years 0.8582 0.8955 0.8701 0.7898 0.7850 0.7775 0.7959
Cities 50–54 years 0.8805 0.9117 0.8921 0.8109 0.8095 0.8043 0.8218
Cities 55–59 years 0.8909 0.9245 0.9098 0.8286 0.8403 0.8350 0.8494
Cities 60–64 years 0.9043 0.9376 0.9264 0.8491 0.8638 0.8580 0.8829
Cities 65–69 years 0.9368 0.9650 0.9571 0.8787 0.8980 0.8911 0.9119
Cities 70–74 years 0.9414 0.9734 0.9718 0.8991 0.9273 0.9253 0.9481
Cities 75 years and older 0.9907 1.0000 0.9961 0.9341 0.9757 0.9700 0.9969
Special Under 1 year 0.8472 1.0000 1.0000 0.9805 1.0000 0.9278 0.8951
Special 1–4 years 0.9881 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
Special 5–14 years 0.6938 0.8084 0.7696 0.7092 0.7141 0.6833 0.7444
Special 15–18 years, Men 0.5542 0.6426 0.6132 0.5302 0.5463 0.5285 0.5724
Special 15–18 years, Women 0.7911 0.9097 0.8697 0.7490 0.7848 0.7445 0.7935
Special 19–44 years, Men 0.6686 0.7325 0.7129 0.6360 0.6412 0.6003 0.6352
Special 19–44 years, Women 0.9422 1.0000 1.0000 0.9068 0.9080 0.8671 0.9237
Special 45–49 years 0.7895 0.8609 0.8397 0.7434 0.7748 0.7365 0.7925
Special 50–54 years 0.8140 0.8797 0.8699 0.7723 0.7854 0.7699 0.8260
Special 55–59 years 0.8475 0.9211 0.9187 0.7947 0.8242 0.8108 0.8619
Special 60–64 years 0.8817 0.9700 0.9719 0.8428 0.8642 0.8480 0.9141
Special 65–69 years 0.9307 1.0000 1.0000 0.8603 0.9032 0.8780 0.9468
Special 70–74 years 0.9347 1.0000 1.0000 0.9212 0.9570 0.9270 0.9928
Special 75 years and older 0.9932 1.0000 1.0000 0.9284 0.9661 0.9628 1.0000
Normal Under 1 year 0.9352 1.0000 1.0000 1.0000 1.0000 0.9802 0.9526
Normal 1–4 years 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
Normal 5–14 years 0.8144 0.8750 0.8406 0.8326 0.8398 0.8105 0.8419
Normal 15–18 years, Men 0.6865 0.7349 0.7046 0.6799 0.6969 0.6767 0.6999
Normal 15–18 years, Women 0.8884 0.9503 0.9075 0.8690 0.8894 0.8484 0.8696

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Espinosa et al. Health Economics Review (2023) 13:15 Page 14 of 20

Table A1 (continued)
Region Age/sex group 2013 2014 2015 2016 2017 2018 2019
Normal 19–44 years, Men 0.7703 0.7781 0.7710 0.7363 0.7244 0.7119 0.7108
Normal 19–44 years, Women 1.0000 1.0000 1.0000 0.9766 0.9696 0.9484 0.9564
Normal 45–49 years 0.8590 0.8744 0.8579 0.8206 0.8228 0.8126 0.8315
Normal 50–54 years 0.8876 0.8974 0.8874 0.8410 0.8454 0.8384 0.8610
Normal 55–59 years 0.9098 0.9211 0.9160 0.8625 0.8753 0.8749 0.8922
Normal 60–64 years 0.9326 0.9510 0.9404 0.8911 0.9071 0.9083 0.9328
Normal 65–69 years 0.9632 0.9906 0.9785 0.9213 0.9415 0.9400 0.9718
Normal 70–74 years 0.9732 0.9960 1.0000 0.9525 0.9702 0.9807 1.0000
Normal 75 years and older 1.0000 1.0000 1.0000 0.9936 1.0000 1.0000 1.0000

Table A2 Values of the frequency adjustment factors for each pricing year
Fit type 2013 → 2015 2014 → 2016 2015 → 2017 2016 → 2018 2017 → 2019 2018 → 2020 2019 → 2021

Effective coverage 1.23% 5.68% 3.62% 0.96% 1.38% 1.93% 1.27%

Disease burden 0.14% 4.97% 7.82% 1.61% 1.44% 8.27% 5.31%

Table A3 Values of the severity adjustment factors for each pricing year
Fit type 2013 → 2015 2014 → 2016 2015 → 2017 2016 → 2018 2017 → 2019 2018 → 2020 2019 → 2021

IBNR 2.61% 2.71% 2.45% 2.69% 3.11% 3.31% 3.74%

Inclusion of technologies 0% 0.09% 0.03% 0.28% 0.76% 0% 0.65%
Comparable 0.34% 0.27% 0.34% 0.35% 0.34% 0.34% 0.01%
Number of attentions 0.27% 5.58% 2.08% 10.96% 3.16% 10.20% 6.76%
Inflation 5.99% 9.18% 11.82% 7.97% 6.86% 6.73% 4.99%

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

 Espinosa et al. Health Economics Review

Table A4 Estimated pure premium – CR-CPU for the remote region (COP)
Effective year 2015 2016 2017 2018 2019 2020 2021
Age/sex group Copula MHSP Copula MHSP Copula MHSP Copula MHSP Copula MHSP Copula MHSP Copula MHSP
(2023) 13:15

Under 1 year 1,527,237 2,320,485 1,508,564 2,539,778 2,105,366 2,748,032 1,842,373 2,963,207 1,544,364 3,120,554 1,028,797 3,287,826 1,055,248 3,458,133
1–4 years 634,899 745,113 582,525 815,527 850,385 882,398 810,835 951,494 743,238 1,002,019 747,229 1,055,728 927,376 1,110,416
5–14 years 299,294 260,282 410,646 284,880 526,003 308,237 424,145 332,372 398,523 350,024 405,041 368,787 444,747 387,886
15–18 years, Men 306,240 248,084 330,218 271,528 564,889 293,793 435,342 316,797 246,661 333,620 683,662 351,504 358,514 369,710
15–18 years, Women 390,085 392,024 447,448 429,073 645,779 464,256 631,040 500,606 622,555 527,190 299,330 555,449 520,348 584,221
19–44 years, Men 407,846 441,437 381,247 483,155 688,408 522,774 627,969 563,708 631,857 593,639 518,382 625,463 656,449 657,859
19–44 years, Women 674,254 818,997 611,898 896,398 990,983 969,897 1,059,223 1,045,843 1,051,801 1,101,380 747,168 1,160,416 1,048,619 1,220,524
45–49 years 779,431 810,084 757,177 886,642 1,249,923 959,345 932,069 1,034,461 1,026,460 1,089,392 791,524 1,147,786 1,161,947 1,207,240
50–54 years 1,014,144 1,033,230 956,565 1,130,873 1,582,293 1,223,599 1,190,627 1,319,412 1,386,178 1,389,471 972,547 1,463,952 1,540,702 1,539,784
55–59 years 1,311,325 1,263,017 1,199,510 1,382,375 2,002,300 1,495,727 1,908,277 1,612,846 1,682,974 1,698,489 1,285,237 1,789,533 1,755,720 1,882,229
60–64 years 1,653,070 1,625,489 1,535,564 1,779,100 2,474,964 1,924,981 2,166,413 2,075,712 2,063,106 2,185,934 1,395,306 2,303,105 2,179,174 2,422,405
65–69 years 2,090,396 2,021,971 1,896,163 2,213,053 3,014,590 2,394,516 3,254,871 2,582,011 3,512,683 2,719,115 1,779,592 2,864,870 2,762,948 3,013,268
70–74 years 2,557,457 2,426,349 2,369,414 2,655,647 3,467,850 2,873,400 4,038,312 3,098,393 3,663,509 3,262,920 2,503,696 3,437,821 3,241,089 3,615,898
75 years and older 3,545,571 3,049,021 3,187,219 3,337,164 4,694,169 3,610,802 4,313,430 3,893,534 3,985,410 4,100,282 2,335,281 4,320,067 4,204,190 4,543,847

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 15 of 20
 Espinosa et al. Health Economics Review

Table A5 Estimated pure premium – CR-CPU for cities region (COP)

Effective year 2015 2016 2017 2018 2019 2020 2021
Age/sex group Copula MHSP Copula MHSP Copula MHSP Copula MHSP Copula MHSP Copula MHSP Copula MHSP
(2023) 13:15

Under 1 year 1,417,639 1,848,647 1,495,897 2,023,351 1,549,862 2,189,255 2,192,416 2,360,680 1,992,175 2,486,033 2,249,336 2,619,291 2,062,813 2,754,962
1–4 years 545,145 593,604 704,723 649,701 605,655 702,976 890,544 758,021 825,014 798,271 896,210 841,062 870,743 884,627
5–14 years 265,372 207,357 476,427 226,952 390,373 245,563 419,110 264,789 400,618 278,851 482,861 293,797 482,894 309,015
15–18 years, Men 273,057 197,640 358,956 216,319 420,098 234,054 404,664 252,383 390,009 265,784 481,441 280,030 464,924 294,535
15–18 years, Women 347,593 312,313 524,442 341,826 483,205 369,856 483,776 398,815 469,554 419,991 541,022 442,506 539,032 465,426
19–44 years, Men 356,061 351,679 474,450 384,912 498,162 416,473 501,367 449,083 497,928 472,933 552,551 498,283 611,970 524,093
19–44 years, Women 594,815 652,468 664,466 714,128 737,620 772,682 850,025 833,185 815,883 877,428 895,888 924,459 925,101 972,347
45–49 years 665,754 645,366 850,475 706,356 916,634 764,274 926,456 824,117 903,457 867,879 1,061,213 914,399 1,087,156 961,765
50–54 years 872,697 823,135 1,117,657 900,924 1,175,257 974,796 1,153,373 1,051,127 1,130,572 1,106,943 1,344,862 1,166,277 1,394,460 1,226,687
55–59 years 1,140,565 1,006,201 1,383,772 1,101,289 1,472,817 1,191,591 1,498,855 1,284,897 1,445,624 1,353,124 1,726,276 1,425,655 1,755,595 1,499,501
60–64 years 1,431,365 1,294,970 1,700,022 1,417,348 1,820,470 1,533,563 1,882,579 1,653,644 1,835,829 1,741,455 2,836,409 1,834,802 2,646,195 1,929,838
65–69 years 1,823,070 1,610,831 2,170,928 1,763,059 2,213,094 1,907,621 2,406,478 2,056,995 2,332,135 2,166,222 2,761,585 2,282,337 2,892,263 2,400,555
70–74 years 2,217,021 1,932,984 2,489,540 2,115,658 2,562,863 2,289,131 2,952,027 2,468,378 2,862,306 2,599,449 3,317,008 2,738,788 3,452,787 2,880,648
75 years and older 3,039,649 2,429,047 3,339,693 2,658,598 3,448,176 2,876,592 3,898,278 3,101,837 3,788,886 3,266,545 4,358,979 3,441,641 4,442,901 3,619,909

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 16 of 20
 Espinosa et al. Health Economics Review

Table A6 Estimated pure premium – CR-CPU for special region (COP)

Effective year 2015 2016 2017 2018 2019 2020 2021
Age/sex group Copula MHSP Copula MHSP Copula MHSP Copula MHSP Copula MHSP Copula MHSP Copula MHSP
(2023) 13:15

Under 1 year 1,022,841 1,851,002 836,705 2,025,927 992,527 2,192,054 1,719,903 2,363,690 1,471,535 2,489,208 2,026,037 2,622,628 1,600,578 2,758,484
1–4 years 381,148 594,362 361,459 650,530 374,824 703,874 598,096 758,986 556,884 799,292 733,951 842,131 719,895 885,754
5–14 years 186,358 207,622 245,202 227,241 260,301 245,877 332,397 265,126 296,648 279,207 424,887 294,173 383,609 309,410
15–18 years, Men 194,820 197,893 194,575 216,594 263,027 234,352 367,272 252,704 345,051 266,124 430,533 280,386 377,955 294,911
15–18 years, Women 244,472 312,709 291,727 342,261 300,648 370,329 491,309 399,324 431,393 420,529 505,268 443,070 453,328 466,022
19–44 years, Men 252,549 352,126 277,259 385,404 317,296 417,007 409,766 449,657 408,279 473,536 449,851 498,918 503,120 524,760
19–44 years, Women 423,303 653,297 376,866 715,036 459,657 773,670 833,239 834,248 799,255 878,549 695,267 925,639 748,503 973,588
45–49 years 479,420 646,189 509,132 707,256 586,840 765,249 799,963 825,170 734,466 868,987 792,222 915,566 825,343 962,993
50–54 years 619,546 824,185 581,631 902,074 749,439 976,044 1,057,708 1,052,465 944,915 1,108,355 968,409 1,167,764 992,632 1,228,255
55–59 years 819,364 1,007,484 765,185 1,102,692 922,316 1,193,114 1,296,407 1,286,533 1,201,790 1,354,851 1,552,970 1,427,473 1,276,633 1,501,416
60–64 years 1,022,272 1,296,619 935,512 1,419,152 1,141,539 1,535,523 1,553,742 1,655,753 1,608,928 1,743,677 1,765,526 1,837,138 1,665,087 1,932,304
65–69 years 1,298,925 1,612,885 1,209,632 1,765,304 1,408,349 1,910,061 1,986,165 2,059,616 2,095,163 2,168,986 2,526,658 2,285,247 2,650,982 2,403,623
70–74 years 1,594,575 1,935,446 1,390,391 2,118,354 1,621,554 2,292,057 2,415,206 2,471,524 2,238,081 2,602,767 3,372,067 2,742,278 2,660,841 2,884,329
75 years and older 2,214,597 2,432,142 1,912,692 2,661,984 2,166,099 2,880,269 3,264,935 3,105,789 3,022,410 3,270,715 3,268,518 3,446,028 3,113,878 3,624,536

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 17 of 20
 Espinosa et al. Health Economics Review

Table A7 Estimated pure premium – CR-CPU for normal region (COP)

Effective year 2015 2016 2017 2018 2019 2020 2021
Age/sex group Copula MHSP Copula MHSP Copula MHSP Copula MHSP Copula MHSP Copula MHSP Copula MHSP
(2023) 13:15

Under 1 year 1,089,246 1,682,733 1,110,173 1,841,752 1,202,462 1,992,772 1,925,328 2,148,810 1,760,026 2,262,913 2,199,322 2,384,209 1,816,719 2,507,708
1–4 years 415,654 540,328 481,795 591,391 477,355 639,884 765,132 689,987 684,052 726,628 783,763 765,576 828,879 805,231
5–14 years 200,316 188,746 379,401 206,582 302,489 223,524 384,539 241,024 354,614 253,825 420,214 267,430 422,288 281,281
15–18 years, Men 206,028 179,901 284,127 196,901 316,979 213,049 388,721 229,729 407,991 241,931 457,309 254,897 451,090 268,100
15–18 years, Women 264,823 284,281 379,881 311,147 363,720 336,662 500,229 363,023 487,110 382,297 514,862 402,790 558,194 423,656
19–44 years, Men 271,091 320,115 335,662 350,367 402,197 379,096 461,455 408,781 448,792 430,486 540,138 453,561 535,278 477,054
19–44 years, Women 446,246 593,908 518,118 650,035 561,305 703,336 849,737 758,406 807,323 798,679 1,033,361 841,490 933,274 885,077
45–49 years 515,979 587,444 628,366 642,959 705,847 695,680 872,501 750,154 825,491 789,987 1,032,640 832,333 1,154,243 875,445
50–54 years 663,234 749,260 825,416 820,067 906,885 887,310 1,088,301 956,788 1,022,411 1,007,595 1,250,903 1,061,602 1,366,329 1,116,591
55–59 years 863,344 915,896 1,020,696 1,002,450 1,130,411 1,084,645 1,415,318 1,169,578 1,326,168 1,231,683 1,595,939 1,297,704 1,977,790 1,364,921
60–64 years 1,099,597 1,178,744 1,259,105 1,290,139 1,408,175 1,395,928 1,729,611 1,505,229 1,688,469 1,585,160 2,268,857 1,670,126 2,652,457 1,756,637
65–69 years 1,399,939 1,466,259 1,583,360 1,604,824 1,745,685 1,736,413 2,233,208 1,872,380 2,155,814 1,971,806 2,529,259 2,077,498 2,621,585 2,185,108
70–74 years 1,721,530 1,759,501 1,864,593 1,925,775 1,970,230 2,083,683 2,790,545 2,246,843 2,604,899 2,366,149 3,018,853 2,492,982 3,170,224 2,622,113
75 years and older 2,366,840 2,211,041 2,465,675 2,419,988 2,663,443 2,618,419 3,482,871 2,823,449 3,339,433 2,973,377 6,168,901 3,132,753 3,597,792 3,295,025

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Page 18 of 20
Espinosa et al. Health Economics Review (2023) 13:15 Page 19 of 20

Abbreviations References
CPU Capitation Payment Unit 1. ADRES. (2020). Ejecución presupuestal URA a corte de 31 de diciembre
SGSSS General System of Social Security in Health (Sistema General de 2020. Retrieved 18 February 2022, from https://​servi​cios.​adres.​gov.​co/​La-​
Seguridad Social en Salud, acronym in Spanish) Entid​ad/​Infor​mación-​finan​ciera/​URA/​Ejecu​ción-​presu​puest​al-​URA
MHSP Ministry of Health and Social Protection 2. Ayyildiz, E., Erdogan, M., & Taskin, A. (2021). Forecasting COVID-19
HBP-CPU Health Benefits Plan recovered cases with Artificial Neural Networks to enable designing an
ADRES SGSSS Resources Administrator (acronym in Spanish) effective blood supply chain. Computers in Biology and Medicine, 139,
EAHBP Entities Administrators of Health Benefit Plans 105029. https://​doi.​org/​10.​1016/j.​compb​iomed.​2021.​105029
COP Colombian pesos 3. Basto S, Bejarano V, Do Nascimento P, Espinosa O, Estrada K, Higuera S, …
CR Contributory regime Barragán L. Producto 3. Estimación actuarial de la UPC del régimen sub-
SR Subsidized regime sidiado. Bogotá: Instituto de Evaluación Tecnológica en Salud y Ministerio
IBNR Incurred but not reported de Hacienda y Crédito Público; 2021.
DANE National Administrative Department of Statistics (Departamento 4. Bolívar, M. (2018). Ajuste de riesgo en la prima de capitación del sistema
Administrativo Nacional de Estadística, acronym in Spanish) de aseguramiento en salud de Colombia para el régimen contributivo
BDUA Database of Affiliates (Base de Datos Única de Afiliados, acronym in (Trabajo Final de Posgrado - Universidad de Buenos Aires). Retrieved from
Spanish) http://​bibli​oteca​digit​al.​econ.​uba.​ar/​downl​oad/​tpos/​1502-​1486_​Boliv​
GD Demand Management (Gestión de la Demanda, acronym in arVar​gasMC.​pdf
Spanish) 5. Cao Q, Leggio K, Schniederjans M. A comparison between Fama and
SISPRO Integrated Social Protection Information System (acronym in French’s model and artificial neural networks in predicting the Chinese
Spanish) stock market. Comput Oper Res. 2005;32(10):2499–512. https://​doi.​org/​
ANN Artificial neural networks 10.​1016/j.​cor.​2004.​03.​015.
MFPC Ministry of Finance and Public Credit 6. Charpentier A, editor. Computational actuarial science with R. Boca
GLM Generalized linear models Raton: Chapman & Hall/CRC; 2015.
MSE Mean square error 7. Cherubini U, Luciano E. Bivariate option pricing with copulas. Applied
MAPE Mean absolute percentage error Mathematical Finance. 2002;9(2):69–85. https://​doi.​org/​10.​1080/​13504​
RSCE Square root of the square differences between the estimated 86021​01367​21a.
copula and empirical copula 8. Chiou, S., & Tsay, R. (2008). A copula-based approach to option pricing
xvCIC Cross-validation copula information criterion and risk assessment. Journal of Data Science, 6, 273–301. Retrieved from
RGOFC Regularized goodness of fit test for copulas https://​www.​jds-​online.​com/​files/​JDS-​503.​pdf
9. Corte Constitucional de Colombia. (2018). Sentencia SU124/18. Retrieved
Acknowledgements 12 February 2022, from https://​www.​corte​const​ituci​onal.​gov.​co/​relat​oria/​
We would like to express our great appreciation to Oscar Melo, Freddy 2018/​SU124-​18.​htm
Hernández, Alejandra Sánchez, Liliana Blanco, Piedad Urdinola, Oscar López, 10. Drummond M, Augustovski F, Bhattacharyya D, Campbell J, Chaiyakanap-
Jaime Ramírez, Pedro do Nascimento, Raúl Macchiavelli, Angélica Ordónez, ruk N, Chen Y, ... Yeung K. Challenges of health technology assessment in
Giancarlo Romano, Esteban Orozco, Sergio Basto, Jhonathan Rodríguez, Diego pluralistic healthcare systems: an ISPOR Council Report. Value in Health.
Ávila, Kelly Estrada, and Juan-Camilo Vargas, for their valuable and constructive 2022;25(8):1257–67. https://​doi.​org/​10.​1016/j.​jval.​2022.​02.​006.
suggestions regarding this research. We are also grateful to the participants at 11. Duarte G, Ozaki V. Pricing crop revenue insurance using parametric copu-
the ‘XXXI International Statistical Symposium’ for their comments and sugges- las. Rev Bras Econ. 2019;73(3):325–43. https://​doi.​org/​10.​5935/​0034-​7140.​
tions. Oscar Espinosa acknowledges Professor Roger Nelsen for his teachings 20190​015.
on the statistical theory of copulas. 12. Duncan I. Healthcare risk adjustment and predictive modelling. 2nd ed.
New Hartford: Actex Publications; 2018.
Authors’ contributions 13. Erdemir Ö, Sucu M. A modified pseudo-copula regression model for
Concept and design: Espinosa, Acquisition of data: Ramos, Martínez, Analysis risk groups with various dependency levels. J Stat Comput Simul.
and interpretation of data: Espinosa, Martínez, Ramos, Bejarano, Drafting of the 2022;92(5):1092–112. https://​doi.​org/​10.​1080/​00949​655.​2021.​19854​98.
manuscript: Espinosa, Martínez, Statistical analysis: Bejarano, Espinosa, Supervi- 14. Escarela D, Carriére J. A bivariate model of claim frequencies and severi-
sion: Espinosa. The author(s) read and approved the final manuscript. ties. J Appl Stat. 2006;33(8):867–83. https://​doi.​org/​10.​1080/​02664​76060​
07439​69.
Funding 15. Frees E, Valdez E. Understanding relationships using copulas. North Am
This research did not receive specific aid from public sector agencies, the Actuarial J. 1998;2(1):1–25. https://​doi.​org/​10.​1080/​10920​277.​1998.​10595​
commercial sector or non-profit entities. 667.
16. Genest, C., Huang, W., & Dufour, J.-M. (2013). A regularized goodness-of-fit
Availability of data and materials test for copulas. Journal de La Société Française de Statistique, 154(1),
Data available on request from the authors. 64–77. Retrieved from https://​jeanm​aried​ufour.​github.​io/​Genest_​Huang_​
Dufour_​2012_​RevFr​ancai​seStat_​Regul​arize​dGood​nessF​itTes​tCopu​las.​pdf
17. Grønneberg S, Hjort N. The copula information criteria. Scand J Stat.
Declarations 2014;41(2):436–59. https://​doi.​org/​10.​1111/​sjos.​12042.
18. Hofert M, Kojadinovic I, Mächler M, Yan J. Elements of copula modeling
Ethics approval and consent to participate with R. Cham: Springer; 2018.
Not applicable. 19. Kholifah A, Lestari D, Devila S. Premium calculation using marginal
generalized linear model combined with copula. AIP Conf Proc.
Consent for publication 2019;2168:020035. https://​doi.​org/​10.​1063/1.​51324​62.
Not applicable. 20. Kularatne T, Li J, Pitt D. On the use of Archimedean copulas for insurance
modelling. Ann Actuarial Sci. 2021;15(1):57–81. https://​doi.​org/​10.​1017/​
Competing interests S1748​49952​00001​47.
None declared by the authors. 21. Lancheros J. Ajuste por riesgo para el cálculo de la UPC en Colombia:
ajuste desde las variables de estado de salud para las aseguradoras
colombianas. 2019. https://​doi.​org/​10.​15332/​tg.​mae.​2020.​0664.
Received: 5 August 2022 Accepted: 23 December 2022 22. Latorre L. Teoría de cópulas. Introducción y aplicaciones a Solvencia II.
Madrid: Fundación MAPFRE; 2017.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Espinosa et al. Health Economics Review (2023) 13:15 Page 20 of 20

23. Lim B, Zohren S. Time-series forecasting with deep learning: a survey. 43. Riascos A, Alfonso E, Romero M. The performance of risk adjustment
Philos Trans A Math Phys Eng Sci. 2021;379(2194):20200209. https://​doi.​ models in Colombian competitive health insurance market. SSRN Elec-
org/​10.​1098/​rsta.​2020.​0209. tron J. 2014. https://​doi.​org/​10.​2139/​ssrn.​24891​83.
24. Mahmoud, A., & Mohammed, A. (2021). A survey on deep learning for 44. Riascos, A., Romero, M., & Serna, N. (2018). Risk adjustment revisited using
time-series forecasting. In A. Hassanien & A. Darwish (Eds.), Machine machine learning techniques. Proceeding Series of the Brazilian Society
learning and big data analytics paradigms: analysis, applications and of Computational and Applied Mathematics, 6(2). https://​doi.​org/​10.​
challenges (pp. 365–392). https://​doi.​org/​10.​1007/​978-3-​030-​59338-4_​19 2139/​ssrn.​29735​14
25. Ministerio de Salud y Protección Social. Estudio de la suficiencia y de los 45. Richman R, Wüthrich M. A neural network extension of the Lee-Carter
mecanismos de ajuste de riesgo para el cálculo de la Unidad de Pago por model to multiple populations. Ann Actuarial Sc. 2021;15(2):346–66.
Capitación para garantizar el Plan Obligatorio de Salud en el año 2006. https://​doi.​org/​10.​1017/​S1748​49951​90000​71.
Bogotá, D.C.; 2006. 46. Shemyakin, A., Zhang, H., Benson, S., Burroughs, R., & Mohr, J. (2019).
26. Ministerio de Salud y Protección Social. Ley 1438 de 2011. Por medio de Copula models of economic capital for life insurance companies.
la cual se reforma el Sistema General de Seguridad Social en Salud y se Retrieved from https://​www.​soa.​org/​globa​lasse​ts/​assets/​files/​resou​rces/​
dictan otras disposiciones. 2011. resea​rch-​report/​2019/​copula-​models-​econo​mic-​capit​al.​pdf
27. Ministerio de Salud y Protección Social. Resolución 5593 de 2015. Por 47. Shi P, Feng X, Boucher J-P. Multilevel modeling of insurance claims using
la cual se fija el valor de la Unidad de Pago por Capitación (UPC) para la copulas. Ann Appl Stat. 2016;10(2):834–63. https://​doi.​org/​10.​1214/​16-​
cobertura del Plan Obligatorio de Salud de los Regímenes Contributivo y AOAS9​14.
Subsidiado para la vigencia 2016 y se dictan otras disposiciones. Bogotá 48. Shi, P., & Lee, G. (2022). Copula regression for compound distributions
D.C.; 2015a. with endogenous covariates with applications in insurance deductible
28. Ministerio de Salud y Protección Social. Resolución 5925. Bogotá pricing. Journal of the American Statistical Association, 1–38. https://​doi.​
D.C.; 2015b. org/​10.​1080/​01621​459.​2022.​20405​19
29. Ministerio de Salud y Protección Social. Resolución 6411 de 2016. Por 49. Shi P, Zhang W. Managed care and health care utilization: specification of
la cual se fija el valor de la Unidad de Pago por Capitación -UPC para la bivariate models using copulas. North Am Actuarial J. 2013;17(4):306–24.
cobertura del Plan de Beneficios en Salud de los Regímenes Contributivo 50. Sklar A. Fonctions de répartition à n dimensions et leurs marges. Pub Inst
y Subsidiado en la vigencia 2017 y se dictan otras disposiciones. Bogotá Statist Univ Paris. 1959;8:229–31.
D.C.; 2016. 51. Tamraz M. Mixture copulas and insurance applications. Ann Actuarial Sci.
30. Ministerio de Salud y Protección Social. Resolución 5268 de 2017. Por 2018;12(2):391–411. https://​doi.​org/​10.​1017/​S1748​49951​80001​2X.
la cual se fija el valor de la Unidad de Pago por Capitación - UPC para el 52. Torres J, Hadjout D, Sebaa A, Martínez-Álvarez F, Troncoso A. Deep learn-
Plan de Beneficios en Salud de los Regímenes Contributivo y Subsidiado ing for time series forecasting: a survey. Big Data. 2021;9(1):3–21. https://​
para la vigencia 2018 y se dictan otras disposiciones. Bogotá D.C.; 2017. doi.​org/​10.​1089/​big.​2020.​0159.
31. Ministerio de Salud y Protección Social. Estudio de suficiencia y de los 53. Werner, G., & Modlin, C. (2016). Basic ratemaking (Vol. 4; Casualty Actuarial
mecanismos de ajuste del riesgo para el cálculo de la Unidad de Pago Society, Ed.).
por Capitación para garantizar el Plan de Beneficios en Salud para el año 54. Xie Y, Yang J, Jiang C, Cai Z, Adagblenya J. Incidence, dependence struc-
2019. In Ministerio de Salud y Protección Social. Bogotá D.C.; 2019a. ture of disease, and rate making for health insurance. Math Probl Eng.
32. Ministerio de Salud y Protección Social. Resolución 3513 de 2019. Por la 2018;2018:1–13. https://​doi.​org/​10.​1155/​2018/​42658​01.
cual se fijan los recursos de la Unidad de Pago por Capitación - UPC para 55. Young, V. (2014). Premium principles. In Wiley StatsRef: Statistics Refer-
financiar los servicios y tecnologías de salud, de los Regímenes Contribu- ence Online. https://​doi.​org/​10.​1002/​97811​18445​112.​stat0​4727
tivo y Subsidiado para la vigencia 2020 y se dictan otras disposiciones. 56. Zhao, X., & Zhou, X. (2012). Estimation of medical costs by copula models
Bogotá D.C.; 2019b. with dynamic change of health status. Insurance: Mathematics and
33. Ministerio de Salud y Protección Social. Estudio de suficiencia y de los Economics, 51(2), 480–491. https://​doi.​org/​10.​1016/j.​insma​theco.​2012.​07.​
mecanismos de ajuste del riesgo para el cálculo de la Unidad de Pago 007
por Capitación, recursos para garantizar la financiación de tecnologías y
servicios de salud en los regímenes Contributivo y Subsidiado. Año 2020.
Bogotá D.C.; 2020a. Publisher’s Note
34. Ministerio de Salud y Protección Social. Resolución 2503. Bogotá Springer Nature remains neutral with regard to jurisdictional claims in pub-
D.C.; 2020b. lished maps and institutional affiliations.
35. Ministerio de Salud y Protección Social. Estudio de suficiencia y de los
mecanismos de ajuste del riesgo para el cálculo de la Unidad de Pago
por Capitación, recursos para garantizar la financiación de tecnologías y
servicios de salud en los regímenes Contributivo y Subsidiado. Año 2021.
In Ministerio de Salud y Protección Social. Bogotá D.C.; 2021a.
36. Ministerio de Salud y Protección Social. Resolución 2381. Bogotá
D.C.; 2021b.
37. Nelsen R. An introduction to copulas (Second Edi). Portland: Springer;
2006.
38. Novales, A. (2017). Cópulas. Retrieved from https://​www.​ucm.​es/​data/​
cont/​media/​www/​pag-​41459/​Copul​as.​pdf
Ready to submit your research ? Choose BMC and benefit from:
39. Oh R, Ahn J, Lee W. On copula-based collective risk models: from elliptical
copulas to vine copulas. Scand Actuar J. 2021;2021(1):1–33. https://​doi.​
• fast, convenient online submission
org/​10.​1080/​03461​238.​2020.​17688​89.
40. Parra, L. (2015). Modelamiento conjunto del número de siniestros y pagos • thorough peer review by experienced researchers in your field
por reclamación en seguros mediante una cópula mixta desde la per- • rapid publication on acceptance
spectiva frecuentista y bayesiana. Retrieved from https://​repos​itorio.​unal.​
• support for research data, including large and complex data types
edu.​co/​bitst​ream/​handle/​unal/​55421/​32297​645_​2015.​pdf?​seque​nce=1
41. Petropoulos F, Apiletti D, Assimakopoulos V, Babai M, Barrow D, Ben Taieb • gold Open Access which fosters wider collaboration and increased citations
S, Ziel F. Forecasting: theory and practice. Int J Forecast. 2022. https://​doi.​ • maximum visibility for your research: over 100M website views per year
org/​10.​1016/j.​ijfor​ecast.​2021.​11.​001.
42. Rao, K., Vecino, A., Roberton, T., López, A., & Noonan, C. (2022). Future At BMC, research is always in progress.
health spending in Latin America and the Caribbean: health expenditure
projections & scenario analysis. Washington, D.C. Learn more biomedcentral.com/submissions

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:

1. use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
2. use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
3. falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
4. use bots or other automated methods to access the content or redirect messages
5. override any security feature or exclusionary protocol; or
6. share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at

[email protected]