Available online at www.worldscientificnews.

com

WSN 153(2) (2021) 157-168 EISSN 2392-2192

Efficiency and Convergence of Bisection, Secant,

and Newton Raphson Methods in Estimating Implied
Volatility

Mahrudinda1,a, Devi Munandar1,2,b, Sri Purwani1,c

1Departement of Mathematic, Faculty of Mathematics and Natural Sciences, Padjadjaran University,
Sumedang, Indonesia
2Research Center for Informatics Department, Indonesian Institute of Sciences, Bandung, Indonesia
a,b,c
E-mail address: [email protected] , [email protected] ,
[email protected]

ABSTRACT
This study aims to estimate volatility prices based on the black-Scholes model (BSM) function
with research data taken during the COVID-19 pandemic. The estimates of the volatility values are
obtained by using three numerical methods, namely the bisection, secant, and Newton Raphson methods.
The numerical processes that produce some iteration results in the three methods are then analyzed and
the best convergence is sought. As a result, Newton Raphson method produces the smallest number of
iterations, which stops at the 3rd iteration and gets a volatility value of 0.500451 with an absolute error
value of 0.000388. However, the method requires an initial approximation which lies only in two
intervals on the axis σ which are close to the true root. Meanwhile, for the other two methods, namely
Bisection and Secant, this limitation does not apply, as long as there is an interval that guarantees the
existence of roots. In this case, bisection method requires11 iterations to converge with volatility value
of 0.500342 and error value of 0.000878. Whereas secant method requires 4 iterations to converge with
a volatility value of 0.500449 and error value of 1.68938E-05. This suggests, that in some cases the use
of Newton method, should be initialized with the use of bisection or secant method, to ensure successful
iteration and accelerate the rate of convergence.

Keywords: Black-Scholes model, volatility, volatility implied, bisection, secant, Newton Raphson

( Received 28 December 2020; Accepted 15 January 2021; Date of Publication 16 January 2021 )
 World Scientific News 153(2) (2021) 157-168

1. INTRODUCTION

Volatility is a statistical measure of fluctuation in stock or foreign exchange prices over

a certain period. High volatility is a price that goes up quickly and then suddenly falls quickly.
The volatility values can be used to estimate the level of risk of loss and even the benefits of
making a decision. There are two kinds of ways to determine the value of volatility, namely
that by using historical data, or that with implied volatility.
According to Canina [1], the value given by implied volatility is better and more
appropriate to use in making a decision option, than the one generated from the calculation of
historical data. This is due to the use of real time data applied by implied volatility. There have
been many studies that describe models for determining implied volatility. The most popular
and widely used one is the black-Scholes model (BSM) published in 1973 [2, 3]. Meanwhile, a
recent research discusses the test of the predictability of the implied volatility of the stock
market using the autoregressive model (AR) [4-6]. Although many new models are available,
we are still interested in using BSM to estimate the implied volatility price.
This is in line with a research emerged in 2020 designing a high-order numerical approach
for efficient solutions of the Black-Scholes fractional equation [7]. This study succeeded in
proving the stability and convergence of BSM. Our study aims to estimate prices of volatility
based on the BSM function with research data taken during the COVID-19 pandemic. The
Covid-19 pandemic has lasted for a long time and at the same time it has given social and
economic impacts widespread.
Monthly production data at US companies show that the COVID-19 shock caused a peak
decline in industrial production by 12-19% [8]. This decline resulted in doubts and affected
stock market players in the world [9, 10]. Hence, the volatility of the stock market experiences
uncertainty in determining market policy. On the other hand, many investors prefer safe-haven
and flight-to-safety behavior.
This means that investors at all the capital markets are more focused on the energy and
precious metal markets, which are not significantly affected by COVID-19 [11]. Based on this,
we tried to use stock data obtained from international precious metal companies to be used as
secondary research data. The numerical methods used to find a solution to the BSM equation
are Bisection, Secant, and Newton Raphson methods, which with their results a comparative
analysis among the three methods was carried out.

2. MATERIALS AND METHODS

2. 1. Materials
This study uses secondary data from the Freeport-McMoRan Inc. (FCX) stock market
obtained at http://www.finance.yahoo.com. The data taken are the current stock price (𝑆0 ), the
strike price (𝐾), and the Call Option Observation price (𝐶𝑜𝑏𝑠 ) with a maturity date (𝑇) of 0.25
years. The share price used is the daily closing share price on November 19, 2020. The interest
rate (𝑟) used is the average World Center Bank interest rate obtained from
https://www.investing.com/
The tool used to find the numerical solutions of the volatility values is software R
Programming, which is also designed to display numerical solutions based on the iterations that
can be shown.

-158-
 World Scientific News 153(2) (2021) 157-168

2. 2. Black-Scholes Function
According to Corbet [12], the option price in the capital market is the same as the
theoretical price calculated using the Black-Scholes formula which can be written as follows,

𝐶𝑜𝑏𝑠 = 𝐶𝑏𝑠 (𝜎), (1)

Meanwhile, the theoretical call option price (𝐶𝑏𝑠 (𝜎)) with volatility (𝜎) from the Black-
Scholes formula can be defined as [2, 11]:

𝐶𝑏𝑠 (𝜎) = 𝑆0 𝑁(𝑑1 ) − 𝐾𝑒 −𝑟(𝑇−𝑡) 𝑁(𝑑2 ), (2)

with 𝑁(𝑑𝑖 ) is a function of the cumulative normal distribution of values 𝑑𝑖 following,

𝑆 𝜎2
𝑙𝑛 ( 𝐾0 ) + (𝑟 + 2 ) (𝑇 − 𝑡) (3)
𝑑1 =
𝜎√𝑇 − 𝑡
𝑑2 = 𝑑1 − 𝜎√𝑇 (4)

Based on equation (1), the BSM function can be formed, namely,

𝑓(𝜎) = 𝐶𝑜𝑏𝑠 − 𝐶𝑏𝑠 (𝜎), (5)

or

𝑓(𝜎) = 𝐶𝑜𝑏𝑠 − [𝑆0 𝑁(𝑑1 ) − 𝐾𝑒 −𝑟(𝑇−𝑡) 𝑁(𝑑2 )], (6)

2. 3. Numerical Methods
The numerical methods used to estimate the volatility value (6) are the bisection method,
the secant method, and the Newton Raphson method. Their algorithms are respectively
presented as follows.

Bisection Method

Step 1: Define the initial approximations 𝜎𝑖−1 and 𝜎𝑖 , also set the error tolerance value
𝜀𝑡𝑜𝑙 = 10−4
Step 2: Calculate 𝑓(𝜎𝑖−1 ) dan 𝑓(𝜎𝑖 )
Step 3: Check whether the function 𝑓 changes sign on an interval [𝜎𝑖−1 , 𝜎𝑖 ], this can be checked
with 𝑓(𝜎𝑖−1 )𝑓(𝜎𝑖 ) < 0. If accepted, the initial approximation values can be used for the
iteration, but if not, select new initial approximation values.
𝜎𝑖−1 +𝜎𝑖
Step 4: Define 𝑐 = 2

Step 5: Calculate the value 𝑓(𝑐)

-159-
 World Scientific News 153(2) (2021) 157-168

Step 6: Perform an evaluation to determine within which subinterval the root of the function
lies. If 𝑓(𝑐)𝑓(𝜎𝑖 ) < 0 then 𝜎𝑖−1 = 𝑐. Otherwise, set 𝜎𝑖 = 𝑐
𝜎𝑖 −𝜎𝑖−1
Step 7: Calculate|𝜀| = 𝜎𝑖

Step 8: Checking, if |𝜀| < 𝜀𝑡𝑜𝑙 with 𝑖 = 1, 2, … , 𝑛 , then the iteration is stopped with 𝑐 as the
estimate of the solution 𝜎 of the volatility function 𝑓(𝜎), but if |𝜀| > 𝜀𝑡𝑜𝑙 , with 𝑖 = 1, 2, … , 𝑛,
then the process is continued back to step 4.

Secant Method

Step 1: Define the initial approximations 𝜎𝑖−1 and 𝜎𝑖 , also set the error tolerance value 𝜀𝑡𝑜𝑙 =
10−4
Step 2: Calculate 𝑓(𝜎𝑖−1 ) and 𝑓(𝜎𝑖 )
𝑓(𝜎𝑖 )(𝜎𝑖 −𝜎𝑖−1 )
Step 3: Calculate the new approximation using 𝜎𝑖+1 = 𝜎𝑖 − 𝑓(𝜎𝑖 )−𝑓(𝜎𝑖−1 )
𝜎𝑖 −𝜎𝑖−1
Step 4: calculate |𝜀| = 𝜎𝑖

Step 5: Checking, if |𝜀| < 𝜀𝑡𝑜𝑙 with 𝑖 = 1, 2, … , 𝑛 , then the iteration is stopped with 𝜎𝑖+1 as
the estimate of the solution 𝜎 of the volatility function 𝑓(𝜎), but if |𝜀| > 𝜀𝑡𝑜𝑙 , with 𝑖 =
1, 2, … , 𝑛, then the process is continued back to step 1 by making 𝜎𝑖 as 𝜎𝑖−1 and 𝜎𝑖+1 as 𝜎𝑖 .

Newton Raphson Method

Derivation of the formula for Newton Raphson method can be obtained geometrically or
with the help of the Taylor series. If 𝜎𝑖−1 is the initial approximation, then the next
approximation can be calculated by the following equation,

𝑓(𝜎𝑖−1 )
𝜎𝑖 = 𝜎𝑖−1 − , 𝑓′(𝜎𝑖−1 ) ≠ 0 (7)
𝑓′(𝜎𝑖−1 )

The derivative of the volatility function (5) can be defined as follows,

𝜕𝐶𝐵𝑆 (𝜎𝑖−1 )
𝑓′(𝜎𝑖−1 ) = − (8)
𝜕𝜎𝑖−1
or,
1 𝑑 2
− 1
𝑓′(𝜎𝑖−1 ) = −𝑆0 √𝑇 − 𝑡 𝑒 2 (9)
√2𝜋

An algorithm for Newton Raphson method is given as follows,

Step 1: Define the initial approximations 𝜎𝑖−1 and the error tolerance value 𝜀𝜀𝑡𝑜𝑙 = 10−4
Step 2: Calculate the value 𝑓(𝜎𝑖−1 ) and 𝑓′(𝜎𝑖−1 )

-160-
 World Scientific News 153(2) (2021) 157-168

Step 3: Determine the next approximate value (7), that is 𝜎𝑖 which lies at the intersection of the
tangent going through (𝜎𝑖−1 , 𝑓(𝜎𝑖−1 )) with axes 𝜎,
𝜎𝑖 −𝜎𝑖−1
Step 4: Calculate |𝜀| = 𝜎𝑖

Step 5: Checking, checking, if |𝜀| < 𝜀𝑡𝑜𝑙 dengan 𝑖 = 1, 2, … , 𝑛 , then the iteration is stopped
with 𝜎𝑖 as the estimate of the solution 𝜎 of the volatility function 𝑓(𝜎), but if |𝜀| > 𝜀𝑡𝑜𝑙 , then
the process is continued back to step 1.

3. RESULTS AND DISCUSSION

Based on the obtained data, the volatility function (6) can be written as:

𝑓(𝜎) = 2,18 − 21,11. 𝑁(𝑑1 ) + 21. 𝑒 −0,0030625𝑁(𝑑2 ) (10)

where,

𝑑1 = 2,0166012𝜎 −1 + 0,25𝜎 (12)

𝑑2 = 2,0166012𝜎 −1 (13)

and
1 𝑑 2
− 1
𝑓′(𝜎) = −21,11 × 0,5 𝑒 2 (14)
√2𝜋

when 𝑓(𝜎) = 0, then it satisfies equation (1) or (6) and has a unique volatility value or values..
This condition can be seen in Figure 1 where the curve intersects with the axis 𝜎 in the
intervals 0 < 𝜎 < 1.

Figure 1. Plot of the volatility function

-161-
 World Scientific News 153(2) (2021) 157-168

The iteration results of estimating the root of 𝑓(𝜎) using the bisection, secant, and
Newton Raphson methods, are respectively shown in Tables 1, 2, and 3. The initial
approximation values used in bisection and tangent methods are 𝜎𝑖−1 = 0.1 and 𝜎𝑖 = 1
respectively, with the tolerance error 0.0001. Whereas, the initial approximation used in
Newton's method is 𝜎𝑖 = 0.1 with the same tolerance error.

Table 1. Iterations of bisection method in finding the volatility value.

𝑖 𝜎𝑖−1 𝑓(𝜎𝑖−1 ) 𝜎𝑖 𝑓(𝜎𝑖 ) 𝜎𝑖+1 𝑓(𝜎𝑖+1 ) |𝜀𝑟 |

1 0.1 1.674629 1 -2.05716 0.55 -0.20603 0.818182

2 0.1 1.674629 0.55 -0.20603 0.325 0.731913 0.692308
3 0.325 0.731913 0.55 -0.20603 0.4375 0.262202 0.257143
4 0.4375 0.262202 0.55 -0.20603 0.49375 0.027881 0.113924
5 0.49375 0.027881 0.55 -0.20603 0.521875 -0.08913 0.053892
6 0.9375 0.027881 0.521875 -0.08913 0.507813 -0.03064 0.027692
7 0.49375 0.027881 0.507813 -0.03064 0.500781 -0.00138 0.014041
8 0.49375 0.027881 0.500781 -0.00138 0.497266 0.013249 0.00707
9 0.497266 0.013249 0.500781 -0.00138 0.499023 0.005934 0.003523
10 0.499023 0.005934 0.500781 -0.00138 0.499902 0.002276 0.001758
11 0.499902 0.002276 0.500781 -0.00138 0.500342 0.000447 0.000878

Figure 2. Volatility value per iteration using bisection method

-162-
 World Scientific News 153(2) (2021) 157-168

It can be seen (see Table 1) that the iteration stops at the 11th iteration and gets a volatility
value of 0.500342 with an absolute error value of 0.000878. Meanwhile, the results of
estimating of the volatility value from the first to the last iteration did not significantly
experience a trend. From the first to the fifth iteration, it is seen that the dynamics of fluctuating
curves are very erratic. The convergence began to appear in the 8th iteration (see Figure 2). The
Bisection method has a linear convergence, the error is reduced by at least ½ of the previous
error. This can be seen in the results shown in Table 1.

Table 2. Iterations of secant method in finding the volatility value.

𝑖 𝜎𝑖−1 𝑓(𝜎𝑖−1 ) 𝜎𝑖 𝑓(𝜎𝑖 ) 𝜎𝑖+1 |𝜀𝑟 |

1 0.1 1.67462931 1 -2.0571645 0.503872 0.9

2 0.503872 -0.0142412 0.1 1.67462931 0.500466 0.403871831
3 0.500466 -7.03E-05 0.503872 -0.0142412 0.500449 0.003405603
4 0.500449 3.8281E-09 0.500466 -7.03E-05 0.500449 1.68938E-05

Based on Table 2, it can be seen that the iteration stops at the 4th iteration and gets a
volatility value of 0.500449 with an absolute error value of 1.68938E-05. Meanwhile, the
results of estimating of the volatility value from the first to the last iteration experienced an
upward trend which immediately converged to the true value (see Figure 3). In this case, the
number of iterations and the absolute error value obtained by the secant method are smaller
than those of the bisection method. This is consistent with the theory that the Secant method
has better convergence rate than the Bisection method, where the rate of convergence for secant
method is of order 1.6.

Figure 3. Volatility value per iteration using secant method

-163-
 World Scientific News 153(2) (2021) 157-168

Table 3. Iterations of newton raphson method in finding the volatility value.

𝑖 𝜎𝑖−1 𝑓(𝜎𝑖−1 ) 𝑓′(𝜎𝑖−1 ) 𝜎𝑖 |𝜀𝑟 |

1 0.1 1.674629 -3.925683 0.526583 0.810096

2 0.526583 -0.108703 -4.129087 0.500257 0.052626
3 0.500257 0.000802 -4.131593 0.500451 0.000388

Based on Table 3, it can be seen that the iteration stops at the 3 rd iteration and gets a
volatility value of 0.500451 with an absolute error value of 0.000388. In this case, the number
of iterations and the absolute error value obtained by the newton method are smaller than those
of the bisection and the secant method. The problem with this method lies in the difficulty of
finding an appropriate initial approximation value which is sometimes close enough to the root.
In addition, Equation (7) requires that 𝑓′(𝜎𝑖−1 ) ≠ 0. Otherwise, is undefined. However, the
Newton method has a quadratic convergence, which means the error decreases by a factor of 2.
This is also shown in Table 3.

Figure 4. The plot of volatility function derivative

Setting the initial value to zero causes the value of 𝜎𝑖 immediately undefined in the first
iteration. Another problem is that there exists an undefined value of 𝜎𝑖 at certain iterations. That
case occurs when the initial approximation used, returns the value of 𝑓′(𝜎𝑖−1 ) being close to
zero, Such examples occur when we use initial approximation values 𝜎𝑖−1 = −10 , and 𝜎𝑖−1 =
0.001. It can be seen from Figures 5 and 6, that the former makes the value of 𝜎𝑖 undefined in
the 3rd iteration, and the latter makes the value of 𝜎𝑖 undefined in the 2nd iteration. This is due
to the value of 𝑓′(𝜎𝑖−1 ) ≈ 0.

-164-
 World Scientific News 153(2) (2021) 157-168

Figure 5. Iterations of the Newton Raphson method with an initial approximation

of 𝝈𝒊−𝟏 = -10

Figure 6. Iterations of the Newton Raphson method with an initial approximation

of 𝝈𝒊−𝟏 = 0.001

By using trial and error operated in the R programming software, we come to the
conclusion that the appropriate initial approximation for the Newton Raphson method lies in
the interval {−5.3233 ≤ 𝜎𝑖−1 ≤-0.0159} or {0,0157 ≤ 𝜎𝑖−1 ≤ 5.8663} (with rounding to 5
digit numbers).
The estimation of volatility value using the Newton Raphson method has taken a smaller
number of iterations. This is due to the rate of convergence of this method is quadratic. Using
𝜎𝑖−1 = 5.8663 as the initial approximation (see Figure 7) Newton method takes 12 iterations
to converge.

-165-
 World Scientific News 153(2) (2021) 157-168

Figure 7. The furthest iteration in understanding the volatility value using the Newton
Raphson method

Very high deviation of fluctuation of volatility values resulted by newton method occurs
in iterations 1-7. In subsequent iterations, the deviation starts to decrease, hence it convergences
to the true volatility value.

4. CONCLUSIONS

The estimates of the volatility values obtained by using the bisection, secant, and Newton
Raphson methods can be seen in Table 4.

Table 4. Volatility approximation value using bisection, secant, and newton raphson method.

Method Total iterations Volatility Absolute error

Bisection 11 50.0342% 0.0878%

Secant 4 50.0449% 1.68938E-03%
Newton Raphson 3 50.0451% 0.0388%

We can see from Table 4, the method with the smallest number of iterations is achieved
by the Newton Raphson method with a total of 3 iterations. Even though the secant method has
the smallest absolute error compared to the other 2 methods, however, with the same number
of iteration, such as 4, Newton Raphson has the smallest absolute error. The problem with the
Newton Raphson method lies in the difficulty of finding an appropriate initial approximation

-166-
 World Scientific News 153(2) (2021) 157-168

that must be in the range {−5.3233 ≤ 𝜎𝑖−1 ≤-0.0159} or {0,0157 ≤ 𝜎𝑖−1 ≤ 5.8663} in this
case. The farther the estimate from the actual volatility value, the more iterations needed.
Therefore, to take advantage of having high rate of convergence owned by the Newton
Raphson method, its use is generally combined with other simple, guaranteed convergence
methods, for example Bisection. Bisection is usually used as initialization providing an
appropriate initial approximation for Newton's method. In this way, hopefully a solution is
guaranteed to exist, and convergence is being faster.
The largest number of iterations that the Newton Raphson method can achieve to calculate
this volatility is 12 iterations. The maximum iteration occurs when the initial estimate used is
𝜎𝑖−1 = 5.8663.

References

[1] L. Canina and S. Figlewski. The informational content of implied volatility. Rev.
Financ. Stud. vol. 6, no. 3, pp. 659–681, 1993
[2] F. Black and M. Scholes. The pricing of options and corporate liabilities. J. Polit. Econ.
vol. 81, no. 3, pp. 637–654, 1973
[3] R. C. Merton. Theory of Rational Option Pricing. Bell J. Econ. Manag. Sci. vol. 4, no.
1, pp. 141–183, 1973. http://www.jstor.org/stable/3003143
[4] Zhifeng Dai, Huiting Zhou, Fenghua Wen, Shaoyi He. Efficient predictability of stock
return volatility: The role of stock market implied volatility. North Am. J. Econ. Financ.
vol. 52, 2020. https://doi.org/10.1016/j.najef.2020.101174
[5] S. Park. The dynamic conditional relationship between stock market returns and implied
volatility. Phys. A Stat. Mech. its Appl. vol. 482, pp. 638–648, 2017, doi:
10.1016/j.physa.2017.04.023
[6] M. McAleer. Asymmetry and leverage in conditional volatility models. Econometrics,
vol. 2, no. 3, pp. 145–150, 2014
[7] P. Roul. A high accuracy numerical method and its convergence for time-fractional
Black-Scholes equation governing European options. Appl. Numer. Math. vol. 151, pp.
472–493, 2020. doi: 10.1016/j.apnum.2019.11.004
[8] D. Altig et al. Economic uncertainty before and during the COVID-19 pandemic. J.
Public Econ. vol. 191, p. 104274, 2020. https://doi.org/10.1016/j.jpubeco.2020.104274
[9] P. K. Narayan, N. Devpura, and H. Wang. Japanese currency and stock market—What
happened during the COVID-19 pandemic? Econ. Anal. Policy, vol. 68, pp. 191–198,
2020. https://doi.org/10.1016/j.eap.2020.09.014
[10] A. A. Salisu and X. V. Vo. Predicting stock returns in the presence of COVID-19
pandemic: The role of health news. Int. Rev. Financ. Anal. vol. 71, p. 101546, 2020,
https://doi.org/10.1016/j.irfa.2020.101546
[11] S. Corbet, G. Hou, Y. Hu, and L. Oxley. The influence of the COVID-19 pandemic on
asset-price discovery: Testing the case of Chinese informational asymmetry.

-167-
 World Scientific News 153(2) (2021) 157-168

International Review of Financial Analysis. 2020 Nov; 72: 101560. doi:

10.1016/j.irfa.2020.101560
[12] R. Baeza-Yates, J. Glaz, H. Gzyl, J. H. Ysler, and J. L. Palacios, Recent advances in
applied probability. Springer Science & Business Media, 2005.
https://doi.org/10.1007/b101429

-168-