Available online at www.sciencedirect.

com

Borsa _Istanbul Review

_
Borsa Istanbul Review 23-3 (2023) 527–540
http://www.elsevier.com/journals/borsa-istanbul-review/2214-8450

Full Length Article

A novel approach to using modern portfolio theory

Taariq G.H. Surtee*, Imhotep Paul Alagidede
Wits Business School, University of the Witwatersrand, Johannesburg, South Africa
Received 13 May 2022; revised 9 December 2022; accepted 9 December 2022
Available online 17 December 2022

Abstract

Since their inception, modern portfolio theory (MPT) and the Sharpe ratio have been among the most popular investment methodologies.
Although MPT has shortcomings, it effectively uses market sentiment to predict low-risk, high-earning portfolios. Our study reviews the current
practice of using the Sharpe ratio or its derivative, the Sortino ratio, and suggests that investors could earn higher returns using Sterling and
Treynor ratios, instead. We find that these two ratios offer higher-performing portfolios, and their statistical distributions have indicators that assist
investors in determining when to use them. These new methods outperform current indexes and funds and are more robust than the capital asset
pricing model used to evaluate investment performance. We conclude by suggesting additional research with different Sterling and Treynor ratios
and advanced optimization algorithms.
Copyright © 2022 Borsa İstanbul Anonim Şirketi. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).

JEL classification: C15; G11; G15; G41

Keywords: Modern portfolio theory; Portfolio screening; Risk-reward ratios

1. Introduction MPT with ratios such as the Sterling and Treynor ratios and
describe our implementation of the models.
This article discusses our novel use of modern portfolio Investors have many ways in which to select portfolios,
theory (MPT) with different risk-reward ratios. Currently, in- including predicting prices based on the news, investor senti-
vestors use MPT with risk-reward ratios such as the Sharpe or ment, and company financial reports. Many other computa-
Sortino ratios, and, at the end of an investment period, they tional optimization models exist, such as the capital asset
evaluate their portfolio performance using ratios such as the pricing model (CAPM), price/earnings ratio models, and mul-
Sterling and Treynor ratios. We pose the question: “Why ticriteria decision models. Investors can employ a combination
would investors use some ratios to propose a portfolio and test of methods depending on their subjectivity. The advantages
the performance with different ratios at the end of the period?” and disadvantages of these models are well documented (see
To answer this question, we look at the Sharpe, Sortino, Basilio, de Freitas, Kämpffe, & Rego, 2018). In this study, we
Sterling, and Treynor ratios and calculate portfolios with each use MPT because it is a popular, well-understood model used
ratio over quarterly intervals for five years. We used the Sharpe in the industry. Moreover, because our goal is to propose using
and Sortino ratios as industry standards in MPT and derived end-of-term ratios such as Sterling and Treynor, MPT offers a
ways to use the Sterling and Treynor ratios similarly to these robust platform for testing performance against the mainstay
standards. Having found positive results, we propose using Sharpe and Sortino ratios. This is the background environment
for explaining a new view of portfolio optimization.
MPT and risk-reward ratios assist investors in evaluating all
* Corresponding author. assets in the same way. This, in turn, allows investors to
E-mail addresses: taariq.surtee@wits.ac.za (T.G.H. Surtee), imhotep.
evaluate many assets together and find portfolios that minimize
alagidede@wits.ac.za (I.P. Alagidede).
Peer review under responsibility of Borsa İstanbul Anonim Şirketi. risk while increasing expected returns. Ideally, investors should

https://doi.org/10.1016/j.bir.2022.12.005
2214-8450/Copyright © 2022 Borsa İstanbul Anonim Şirketi. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://
creativecommons.org/licenses/by-nc-nd/4.0/).
T.G.H. Surtee, I.P. Alagidede _
Borsa Istanbul Review 23-3 (2023) 527–540

understand as many elements of risk as possible. The elements only sharia-compliant assets and limiting optimization to long-
of risk are embedded in the many data sources available to only portfolios.
investors (news, sentiment, financial reports). Therefore, using We further test the robustness of our models against an
one method, such as interpreting the news, might lead to industry-standard CAPM. We found that we could improve
overlooking essential risk information from other sources. portfolio performance by applying different ratios’ predictions
Hence, investors who base their portfolios only on the news depending on market conditions. We also found that ratios like
may not have the same views as investors who base their the Treynor ratio predict different portfolios from the other
portfolios on financial reports. ratios. Treynor is beta focused, which could explain the phe-
Moreover, several investors might look at the same data and nomenon. This observation is explored here and offers further
have different views based on their choice of sources, invest- research opportunities.
ment horizons, strategies, or personal subjectivity. Hence, The remaining sections develop the background models,
having a method that considers all investors' views (market theoretical basis and interpretation of the risk-reward ratios
sentiment) is advantageous. MPT (using price volatility as risk) used to create our novel approach. The article concludes with
is indifferent to the number, weights, or types of risk variables our findings and further research suggestions. Section 2 ex-
used in other models. Furthermore, because it considers price plores the current industry MPT use describing its strengths
volatility, MPT inherently incorporates the net effect of market and weaknesses. Section 3 develops the model building in our
sentiment. In this way, MPT has the broadest view of what interpretation. Section 4 offers details of the risk-reward ra-
investors think about the market. tios calculations and how we used the data. Section 5 dis-
However, MPT is not without fault. Although Sharpe (1994) cusses our findings, focusing on unique scenarios and offers
considers this approach a complete view of an asset's perfor- future research opportunities. We conclude our article in
mance, it uses historical data in proposing a portfolio, which section 6.
could be problematic (Rice, 2017, pp. 13–18; Schulmerich,
Leporcher, & Eu, 2015). MPT theory and Sharpe assume that 2. Background: modern portfolio theory
investors' views and goals are homogeneous. However, MPT
might not account for recent market activity. Further, investors Modern portfolio theory (MPT) and the Sharpe ratio are
may not behave in the same way, resulting in different portfo- investment tools widely used in industry (Contreras, Lizama, &
lios. Hence, the suggestions by MPT could be incorrect. Stein, 2016, pp. 1–25). Harry Markowitz (who created MPT)
Additionally, because MPT uses historical data, it does not and William F. Sharpe (the Sharpe ratio) are Nobel prize
forecast future events. Some events become known, and market winners. Although the Literature does not indicate why their
sentiment corrects for them, but others remain unknown until theories are the mainstay of industry, the fact that they won
they severely affect the market. Investors must be aware that Nobel prizes might suggest why these models are used
this could happen and affect their portfolios. Further, when together. Investors could use subjectivity and sentiment when
used with the Sharpe ratio, MPT is criticized for its assumption picking methods and theories, and the attention from the Nobel
that volatility is normally distributed (see Hawley & prizes won by these economists might influence investor
Lukomnik, 2018; Schulmerich et al., 2015). These additional choice. We argue that MPT is a comprehensive analysis tool
assumptions have been tested in many studies in which the and is suitable as an introduction to our novel approach. After
effects of different statistical distributions such as skew or the fundamentals of the approach are understood, other opti-
kurtotic distributions are evaluated (see Elton & Gruber, 1997; mization tools can be applied.
Schulmerich et al., 2015). MPT uses utility and risk to determine the most efficient
Other studies simulate distributions, finding that negative investment portfolios by changing the asset weights (ωi ), to
skewness and leptokurtic (positive kurtosis) distributions are create a portfolio whose total weight is one. The portfolio
favorable (Xiong & Idzorek, 2011). Investors may find superior equation is E(Rp ) = ∑ni=1 ωi E(Ri ), where E(Rp ) = 1. Applying
gains with negatively skewed distributions, which lead to MPT with different risk-return ratios results in different effi-
returns above the average, whereas positive kurtotic distribu- cient frontiers for the same assets. MPT uses covariance be-
tions offer lower risk. Xiong and Idzorek (2011) find that tests tween assets when calculating an efficient frontier and the
with different distributions outperform MPT with standard associated optimal portfolio (Contreras et al., 2016, pp. 1–25;
assumptions in some cases. Further studies test various hy- Markowitz, 1952).
potheses, showing that investors have different return horizons The ratios interpret risk and reward differently, influencing
(Hawley & Lukomnik, 2018). They find that homogeneous how an investor's portfolio might grow (their alphas) with
expectations can be disregarded, but MPT still yields effi- respect to the market (Goetzmann, Ingersoll, Spiegel, & Welch,
ciency. We review our results under these criteria to establish 2002; Rollinger & Hoffman, 2013). This is evident in the shape
the performance of our models. of the MPT efficient frontiers produced by the ratios. When the
This article is part of a more extensive study conducted to degree of difference in the shape of a portfolio's efficient
improve the performance of sharia-compliant investments in frontiers from that of the market is greater, the portfolio shows
South Africa. As such, we compare our performance measure greater independence.
against the South African Sharia Index. Further, to accommo- The shape of efficient frontiers represents the risk for all
date sharia requirements, we constrain the MPT model, adding possible portfolios. The slope between the market proxy
528
T.G.H. Surtee, I.P. Alagidede _
Borsa Istanbul Review 23-3 (2023) 527–540

(hurdle rate) and the efficient frontier identifies the optimal derivations of our ratios can offer new insights into portfolio
portfolio. The slope indicates risk fluctuation sensitivity, in dynamics and better performance.
which a steeper line indicates more volatility in returns within a
risk band. Further, if the market proxy changes, the slope also 3. Theoretical literature review
changes. Hence, the risk sensitivity and optimal portfolio
change, with the corresponding alphas and betas of the ex- Because MPT is a theoretical model, it makes several as-
pected returns (Uppal & Zaffaroni, 2016). Ideally, the market's sumptions. Some assumptions hold in our constrained model
efficient frontier is beta. Any change in the shape of the port- discussed here, while others might not.2 MPT is a model
folio's efficient frontier could be attributed to alpha. Any designed to provide perpetual growth while minimizing risk
portfolio not on an efficiency frontier implies that for a set of (Markowitz, 1952).
assets, either more returns could be earned for the same risk or The MPT model is defined as:minimize ∑ni=1 ∑nj=1 ωi ωj σ ij
the risk itself could be lower, meaning the portfolios could be (risk), and maximize the expected return,
inefficient. n
Different ratios for the same set of assets provide different E(Rp ) = ∑j=1 ωi E(Rp ),
efficient frontiers and portfolios.1 The slope of the tangent lines
shows how the returns' sensitivity varies for changes in risk. where:
Because each ratio's risk (denominator of the ratio) is inter- Rx is the expected return (simplified notation for E(Rp ) as
preted differently, a direct comparison may not make sense. noted above)
Because it is possible to calculate all four ratios for each Rp is the return on the portfolio p,
portfolio, we can find a common comparison platform and Ri is the return on asset i
compare the results of all the models. ωi is the weight ω of component asset i
However, MPT has some inefficiencies and faults. Although σ ij is the covariance between assets i and j (Markowitz,
Sharpe (1994) considers this approach a complete view of an 1952).
asset's performance, it uses historical data in proposing a The model assumes that all investors have homogeneous
portfolio, which could be problematic (Rice, 2017, pp. 13–18; expectations and that this will not change over time. Risk is
Schulmerich et al., 2015). MPT theory and Sharpe assume that expressed statistically as volatility, and efficiency is calculated
investors agree that future returns are based on historical per- with the shared interaction (covariance, σ ij ) between assets.
formance and that investor outcomes are all the same (homo- However, some theoretical assumptions exist when interpreting
geneous). However, MPT might not account for recent activity market sentiment as volatility. The assumptions imply that
(e.g., breaking news). investors’ outcomes are homogeneous, and distributions have
Further, when used with the Sharpe ratio, MPT is criticized probabilistic returns with the lowest normally distributed
due to its assumption that the volatility is normally distributed volatility (Hawley & Lukomnik, 2018; Rice, 2017, pp. 13–18;
and that market sentiment can be assumed (e.g., Hawley & Schulmerich et al., 2015). Ideally, the assumptions are that all
Lukomnik, 2018; Schulmerich et al., 2015). Many studies investors calculate a risk-return portfolio for a single period
have tested different statistical distributions such as skew or (for all time).
kurtotic distributions (e.g., Elton & Gruber, 1997; Schulmerich Further, it assumes that volatility is normally distributed,
et al., 2015), showing changes in expected outcomes. Further and we can make statistical inferences assuming that the dis-
studies have simulated distributions, finding that negatively tributions are normal. However, historical data show that the
skewed and leptokurtic distributions are beneficial (Xiong & distributions might not be normal and that market sentiment
Idzorek, 2011). This makes sense because investors can ach- changes over time (Hawley & Lukomnik, 2017). Additionally,
ieve higher gains with negatively skewed distributions since some investors create portfolios such as hedge funds with
the distribution favors above-average returns. Positive kurtotic highly skewed or kurtotic data (Davies, Kat, & Lu, 2016).
distributions suggest that expected returns variance is lower Hence, market sentiment could influence the appearance of the
than normally distributed returns. Xiong and Idzorek (2011) distributions of the assets. Studies have shown that the effects
find that the tests with different distributions outperform of correcting for normal distributions do not always yield
MPT with standard assumptions. However, their studies use significantly better results (Elton & Gruber, 1997; Schulmerich
unconstrained models. Hence, we test their findings in our et al., 2015). Given that, investors who rely on MPT might
constrained models. subjectively choose to review the shape of the distributions
Moreover, the Sharpe ratio's strengths and weaknesses are
thoroughly discussed in Section 4. Although they indicate
varied results in the portfolios proposed, many studies do not
2
significantly raise the results over an extended period. This was Ordinarily, in the purest form, investors input data for all investment assets
not the case using our novel approach, suggesting that using into MPT in order to determine the portfolio. MPT assumes perpetual growth
and includes short selling. However, we base our constraints on Sharia re-
quirements, using three main criteria. First, we limit assets to those that fit
Islamic ideology; second, we rebalance the portfolio if an asset became non-
1
Investors could also obtain different efficient frontiers using the same ratio compliant with Islam; and, third, because Islam prohibits short selling, we
in an unconstrained or differently constrained MPT model. constrain the model to long-only portfolios.

529
T.G.H. Surtee, I.P. Alagidede _
Borsa Istanbul Review 23-3 (2023) 527–540

when picking portfolios. We discuss the shape of the distri- (CPI + 2.5%). We also use a rate that, at a minimum, equals
butions in Section 5.2 and find that they can enhance inflation after subtracting zakat.
prediction. Preliminarily, because we looked at sharia investment in
In considering data, some studies show that high-frequency South Africa in a broader study, we compare our results to
risk data can increase performance (Ma, Li, Liu, & Zhang, those with the South African Sharia Index (SASI). Fig. 1 shows
2018). We chose to use daily data. There are approximately that, although it slightly underperforms, the SASI is compa-
eighty daily and sixteen weekly data points when rebalancing rable to the South African FSTE All-Share index. The graph
quarterly. Setting the assumption of a normal distribution aside plots the performance of South Africa's major indexes and our
for a moment, daily data maximize the size of the data set and best-performing portfolio. The SASI (yellow line) performs
provide more accurate distributions. In that way, more data similarly to the Top 40 and All-Share indexes, albeit generally
gives us a more precise view with which to analyze the lower, as shown from June 2015 to September 2019. Our best-
normality assumption when determining the skewness and performing portfolio (the green line) consistently outperforms
kurtosis of the distribution. However, investors do not have all the indexes throughout the period. The statistical and eco-
homogeneous expectations and might not follow suit. nomic significance in the results is discussed in Sections 5.2
Further, assets can be traded at different frequencies, mak- and 5.3, respectively, in the findings chapter.
ing the optimal analysis frequency dynamic. Hence, the anal- In sum, although MPT remains one of the tools at the
ysis done with this frequency could differ from analysis done disposal of investors, it is governed by assumptions and has
with other frequencies. Therefore, after our approach is un- material inefficiencies and human bias. Some studies claim that
derstood, research in more comprehensive optimization algo- sentiment models (in behavioral economics) are more adept at
rithms can be used to test different frequencies in the assets and incorporating human bias (e.g., Simo-Kengne, Ababio, Mba, &
the MPT analysis. Koumba, 2018). However, we argue that they are implied in
Moreover, our study is based on sharia assumptions. the net effect of volatility. Like most models, MPT fails to use
Therefore, our interpretation of the MPT differs from other the latest market information effectively or to make speculative
studies in three ways. First, we do not simulate short selling in assumptions that could cause unpredictable behavior. Noting
the market. Second, we limit assets to those that are sharia these limitations, we show that using MPT can offer high-
compliant, and, third, we use the South African Sharia Index performance portfolios. To use MPT as an investment tool,
(SASI) instead of the standard market proxy used by other investors must accept the principles and assumptions on which
investors. Fourth, unlike other studies, we deduct the annual MPT is built. That is, all price changes are reflected in the
compulsory religious alms called zakat when reporting results. volatility (Markowitz, 1952). It is up to the investor to consider
The Literature discusses using the risk-free interest rate as a what affects an asset's price volatility. If investors can accept
common starting point when calculating an optimal portfolio. these assumptions, they should be able to rationalize the use of
This homogeneous expectation does not directly apply to Islam risk-reward ratios and MPT. Further, MPT allows investors to
since Islam outright rejects interest. Instead, we use the con- relax constraints that may depart from homogeneity and could
sumer price index (CPI) with zakat (2.5%) as our hurdle rate affect performance.

Fig. 1. Comparison of the performance by South African indexes and the best portfolio.

530
T.G.H. Surtee, I.P. Alagidede _
Borsa Istanbul Review 23-3 (2023) 527–540

4. MPT ratios and variables Sortino ratio, which uses drawdown data for the denominator.
However, the numerator or expected return is interpreted
Originally, MPT was devised as an economic theory to differently. The Sterling ratio used in this study is adapted
optimize the possible risk of any additional utility earned. from:
Moreover, MPT with risk-return ratios is the most popular
CARR
quantitative investment method used. The risk-return ratio Sterling ratio =
terminology is commonly understood. Typically, returns are |ALD|′
calculated as the returns of an asset minus the risk-free returns
where:
(Rx − Rf ). However, we use ln(pt /pt−1 ) to give a relative view
CARR is the compound annualized rate of return.
of change.
ALD is the average largest drawdown (Magdon-Ismail &
Atiya, 2004)
4.1. Sharpe and Sortino Ratio
The ratio proposes using compound annual rates of return
(CARR), instead of ln(pt /pt−1 ) . The risk is measured as the
The most commonly used ratios are constructed as follows:
average largest drawdown (ALD). In some cases, 10 percent is
ln(pt /pt−1 ) subtracted from the ALD to account for bond yield rates.
Sharpe Ratio = , Sterling proposed this when the ratio was defined (Kolbadi &
σx
Ahmadinia, 2011), but it is optional since we use a hurdle
where: rate. Further, the original period was defined as one year.
ln(pt /pt−1 ) is the expected return However, the periods used in the numerator and denominator
σ x is the standard deviation of the expected return (Sharpe, are flexible as long as both are calculated within a fixed period
1994) of analysis (Kolbadi & Ahmadinia, 2011; Magdon-Ismail &
The Sortino ratio is different from the Sharpe ratio in that it Atiya, 2004).
only uses downside standard deviation. No research detailing the use of the Sterling ratio in MPT
could be found. If we calculated CARR using the entire period,
ln(pt /pt−1 ) we could not calculate covariance. Therefore, we adapted the
Sortino Ratio = ,
drawdown(σ x ) ratio in two ways to work in MPT. First, because we rebalance
the portfolios, we calculate CARR over monthly periods, such
where: that, for a one-year period of analysis, there are twelve CARR
ln(pt /pt−1 ) is the expected return values. Second, when calculating the ratio, we use daily ALD.
drawdown(σ x ) is the standard deviation of negative returns We use daily drawdowns because we include daily data in our
(Rollinger & Hoffman, 2013) analysis and the previous two ratios. This frequency aligns with
The Sharpe and Sortino ratios are well known and written the finding that high-frequency risk data supports better port-
about widely. Their strengths and weaknesses are the subject of folio performance, and testing different frequency data is rec-
research from their introduction to date (e.g., Goetzmann et al., ommended to determine efficiency (Ma et al., 2018). With
2002; Kaplanski, Levy, Veld, & Veld-Merkoulova, 2016; these changes to the Sterling ratio, we could calculate
Rollinger & Hoffman, 2013; Scholz, 2007; Sharpe, 1994). covariance.
To implement the ratios and MPT simulations, we use Hence, the difference between the Sterling ratio and the
Microsoft Excel (Excel). The numerator in this study is the Sortino ratio is in the numerator in which the analysis uses
average daily returns earnings net of the hurdle rate.3 Average CARR. Interpreting the ratio in this way maintains the purpose
pricet−1 )), and t
closing pricet of the Sterling ratio. CARR smooths out the returns and ex-
daily returns are calculated as average(ln(closing
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ presses the expected returns as a compound investment return.
∑ (cpt −average(cp))2
is the date. The denominator is , for all t in
end price /(t*frequency)
1
count(cp)
the Sharpe ratio, and cpt is capped at zero for the Sortino ratio, CARR = frequency(( ) − 1),
start price
so only negative volatility is considered. The advantages and
disadvantages of Excel are discussed in Section 6, which dis- where t is the number of years.
cusses the algorithms. The Treynor ratio was defined to evaluate a portfolio's
The Sterling ratio is ordinarily used at the end of a period performance against market risk (Hübner, 2005). Ideally, the
and not used in MPT. In our study, we reinterpreted the ratio to market has a specific Treynor ratio over a period, and the
use it as a periodic risk-reward ratio. There are several in- Treynor ratio measures a portfolio's susceptibility to changes in
terpretations of the Sterling ratio (e.g., Kolbadi & Ahmadinia, the market. As with the other ratios, if an investor's portfolio
2011; Magdon-Ismail & Atiya, 2004). This ratio is like the has a higher ratio, it could outperform the market for the same
risk or conversely earn equivalent returns while reducing the
effect of market risk (market beta risk). Beta is calculated as
3
Daily is chosen because it is convenient for creating large data sets. It is a ,Rm )
easier to determine distribution normality with larger data sets. Daily could
βa = cov(R
var(Rm ) , where Ra is the returns on assets, and Rm is the
have been any subjective period like a weekly period. market returns. When the Treynor ratio is calculated, each
531
T.G.H. Surtee, I.P. Alagidede _
Borsa Istanbul Review 23-3 (2023) 527–540

asset's beta is calculated, and the weighted portfolio has the Table 1
combined beta risk (βx ) of all the assets. Each asset's beta may List of variables derived and where they are used.
be unique and contribute uniquely to the portfolio's beta, Sharpe Sortino Sterling Treynor
depending on their weight. Beta is calculated using both upside Returns (Numerator)
and drawdown volatility, so the interpretation suffers from the Expected returns X X X
same issues as the Sharpe ratio, in which upside volatility may Compound annualized rate of return X
Risk (Denominator)
increase total volatility. Therefore, further studies should look Standard deviation X
at a drawdown equivalent to beta. Similarly, ratios that use Drawdown standard deviation X X
alpha should also be considered in further studies. Beta of the expected return X
For this study, the Treynor ratio uses the standard beta
interpretation. The ratio is:
4.2. Data collection, cleaning, and derivation
ln(pt /pt−1 )
Treynor Ratio = ,
βx The data source for the ratios is daily price data collected
from the Johannesburg Stock Exchange (JSE Marketing, 2013)
where: or Yahoo Finance (Yahoo, 2021). All assets were collated in a
Rx is the expected return single Excel spreadsheet. Missing data were imputed as an
Rf is the risk-free rate (market proxy) average using the day before and after. If there were a few days
βx is the portfolio's beta of the expected return (Hübner, of missing data, each missing data point was calculated
2005) incrementally (or decrementally) from the day before. We
The Treynor ratio also needs to be interpreted for use in excluded the asset if any data was missing more than 5 data
MPT because betas are generally calculated over an entire points. The imputation calculation used is:
period. We calculate betas at monthly intervals over an in- cpn − cp1
vestment period to derive the covariance of betas. Calculating cpt+1 = cpt + , for all missing data, m,
n−1
betas for a weighted portfolio is a common practice (Pettengill,
Sundaram, & Mathur, 1995). However, no evidence could be where t = 1. m and n = (m + 2)
found to calculate (up to 12 in our case) betas within a period. All the data were converted to South African cents (cur-
In using the Treynor ratio, our study optimizes beta-risk rency), and the returns were normalized. The initial total
ln(pt /pt−1 )
adjusted returns, in which our Treynor ratio is covariance (βx ) . portfolio was made 100 cents. Thereafter, because MPT relies
The critical difference in the Treynor ratio is that it uses beta on complete historical data during each period of analysis, only
(systemic risk). By minimizing only beta, the MPT model al- assets for which all the data were available for that period were
lows any level of alpha risk. Generally, performance yields used. Further, because we perform long-only constrained MPT
vary in different market scenarios (Ashraf, 2013; Ashraf & analysis, the data were screened for assets with positive returns
Mohammad, 2014; Masih, Kamil, & Bacha, 2018). Without (( p end date) – (p start date) > 0) to manage the size of the
knowing the alpha, investors may take unacceptable risks that data set.
they would not have taken after reviewing the total risk.
Alternatively, if assets are highly correlated to beta, Treynor's 5. Analysis
results will be similar to those of the index (South Africa Sharia
Index in our case), yielding similar results. Therefore, we We use three months and stock exchange news to confirm
calculate all 4 ratios for all 19 portfolio iterations and compare whether assets remain sharia compliant and to rebalance the
the results to understand the findings further. portfolio. Hence, we rebalance the portfolios every three
Each ratio is subject to the same criticism, namely, that it months over five years (starting January 2, 2015). Investors can
implies that previous performance predicts future performance. choose to rebalance at different intervals. The first three itera-
Investors may find that recent activity is not reflected accurately tions have less data; after that, we use 252 trading days. In
in forecasting performance. However, although these criticisms some cases, less than 252 trading days remain in the calendar
are valid, the ratios have particular merit in quantifying portfolio year. To compensate, we use data from the previous December.
performance and considering market sentiment to give investors
insight into how other investors may view assets. The differ-
ences between the ratios are in how returns and risk are inter- Table 2
preted. The Sharpe, Sortino, and Treynor ratios consider Number of data points per period.
average growth over a period, whereas the Sterling ratio con- Iteration Period length Analysis Period Number of
siders a compound return. Both can be used to calculate returns data points
over future periods. Further, each ratio has a unique interpre- 1 Three months January to March 2015 63
tation of risk. These differences are illustrated in Table 1, which 2 Six months January to June 2015 123
shows the variables used for calculating the four ratios. The 3 Nine months January to September 2015 187
4 One year January to December 2015 252
numerator is derived as a form of returns for each ratio, and the
denominator is derived as some form of associated risk. Thereafter, we used 252 of the latest data points.

532
T.G.H. Surtee, I.P. Alagidede _
Borsa Istanbul Review 23-3 (2023) 527–540

Fig. 2. Depiction of how data were used to model portfolios.

Table 2 shows the periods used and the number of data points next section. Additionally, we analyze the distributions for the
per period. We had 63 data points for the first iteration, 123 for next period to reveal predictive trends. We then compare our
the second, 187 for the third and 252 data points thereafter. robustness results to those from the CAPM. The CAPM is
The simulations are repeated every three months, and the calculated using various coefficients, and the same betas
portfolios are amended with the new suggested assets. The calculated for the Treynor ratio. We benchmark the errors of
results are used to derive returns for the next three months. The our MPT models and the CAPM with the same data to discuss
results of the three months going forward are taken as actual the statistical and economic significance. Our findings are
returns that the portfolio would have yielded. The analysis and discussed in the next section.
rebalance timeline are illustrated in Fig. 2.
The preliminary tests include comparing the various MPT 6. Findings
models and current South African indexes and products. A
difference of two means t-tests is used to evaluate the models’ Simulating performance with historical returns does not
expected returns and volatility (DeFusco, McLeavey, Anson, guarantee future returns. However, they offer a way to under-
Pinto, & Runkle, 2015). Because we use historical data in stand the behavior of assets and portfolios to assist investors in
the study, we can evaluate our results with actual data in the decision-making (screening). After each simulation, we select
following period, answering the performance evaluation the results that seem to fit our highest returns and lowest risk
questions proposed. drawdown criteria. We find that we did not always choose the
Each ratio predicts different weighted portfolios. Because best options. Table 3 presents the summary statistics for the
the volatility interpretation for each ratio is different, a straight analysis. The last two rows tabulate the results of our subjec-
comparison of the model's portfolios is unfounded. Therefore, tively chosen portfolio and the best possible portfolio we could
all the ratios are calculated for all iterations, allowing us to have found based on our simulations. The best returns portfolio
analyze the risk and results for all the models. We obtain assumes that an investor could always select the best-
efficient frontier curves, tangent slopes, and loss probability performing portfolio after each iteration.
analysis. The “Final Returns” column shows real growth from July 1,
After we complete our simulations, we discuss our findings 2015, to December 31, 2019. Except for Treynor, our MPT
and compare them to the established indexes and funds in the simulations outperform the current funds and indexes. The

Table 3
Summarized comparison of results at the end of the period of analysis.
Final returns Net return Std Dev (Risk) Std Dev DD Sharpe ratio for the Sortino ratio for the
after zakat and CPI (Drawdown risk) index or portfolio index or portfolio
Current indexes and funds:
SASI 118.68% −10.38% 1.14% 0.75% −9.14 −13.82
AltX 80.96% −48.10% 1.12% 0.84% −42.83 −57.36
ABSA 93.59% −35.47% 2.63% 1.84% −13.48 −19.24
Kagiso 128.94% −0.12% 0.65% 0.42% −0.19 −0.28
Our MPT simulations:
Sharpe 146.40% 17.33% 0.88% 0.59% 19.61 29.49
Sortino 149.24% 20.17% 0.94% 0.61% 21.37 32.81
Sterling 159.26% 30.20% 1.01% 0.65% 29.86 46.55
Treynor 114.99% −14.07% 1.06% 0.69% −13.22 −20.24
Subjective 133.60% 4.53% 0.91% 0.60% 4.99 7.52
Best Returns 201.66% 72.60% 0.74% 0.47% 97.82 153.14
Note: Red indicates negative returns against a five-year inflation-adjusted increase. The inflation-adjusted value of 100 cents is R1,2906 at the end of the analysis
period.

533
T.G.H. Surtee, I.P. Alagidede _
Borsa Istanbul Review 23-3 (2023) 527–540

Table 4 and Sterling three times. However, if we had had better fore-
Summarized comparison of results. sight, we could have selected ratios that yield the Best Returns
Subjective Returns Best Returns portfolio. As in Table 4, the Best Returns portfolio primarily
Sharpe 6 4 has Sterling and Treynor predictions, meaning other screening
Sortino 10 1 tools are needed. The Literature suggests that statistical dis-
Sterling 3 6 tributions can infer portfolio suggestions. Therefore, we
Treynor 0 8
Total iterations 19 19
perform an in-depth analysis of the results and their statistical
information to understand the results of the ratios. We find that
investors could trade off slightly higher risk and select the
Sterling ratio. However, it was not so easy with the Treynor
“Net Returns after zakat and CPI” column shows that none of predictions.
the existing funds and indexes keep up with CPI, whereas, in
general, our models do. Additionally, we correct for religious 6.1. Understanding Treynor results
alms (zakat) over the 5 years; hence, our comparative returns
are higher. The Treynor ratio yields higher returns in 8 of the 19 iter-
The “Risk” and “Drawdown Risk” columns show that our ations. Unlike Treynor's predictions, Sharpe ratios are consis-
risk profiles are generally lower than the current alternatives, tently lower than the others, and the drawdown risk is
except for Kagiso. However, MPT typically has higher returns consistently higher; hence, it is never picked. Moreover, the
for that risk. The last two columns show that the industry uses statistics show that the Treynor model is the most volatile.
Sharpe and Sortino ratios. Even though Kagiso may be a viable Fig. 3 shows that Treynor (gray line) predicts better returns
fund, it does not add value to the investor because its returns with more data. In 2015, with less data, it predicted the worst-
are lower than our hurdle rate. The “Best Returns” row with a performing portfolios twice. In 2016 Treynor was the “Best
101.66 return shows that investors could have earned 3.5 times Returns” portfolio choice. However, in 2019, when all the
the returns of Kagiso for similar risk. Hence, we find that our other ratios predict high-performing portfolios, Treynor pre-
models offer better performance, and we continue to discuss dicts lower-performing portfolios. Treynor asset weights are
how to select portfolios to improve returns. Net return after different from those of the other ratios. This contributes to
zakat and CPI are a magnitude larger since the Kagiso fund Treynor's unique performance and higher volatility. For
almost breaks even while the best portfolio returns 72.6 example, in some periods Treynor predicts the worst-
percent. performing portfolio, and the other ratios also perform
Table 4 presents our choices for obtaining a Subjective or poorly. However, the portfolio predicted by Treynor recovers
Best Returns portfolio as tabulated in the last two rows of Table entirely in the next period, but the other ratios do not. Table 5
3. For Subjective Returns, we constantly chose ratios with shows the sample data and returns over these periods.
lower drawdown volatility, and when the drawdown was Although the values are not important, the difference in
similar, we picked higher returns. Hence, in the nineteen iter- portfolio weight choices and the results are. It appears that
ations, we chose the Sharpe ratio six times, Sortino ten times, Treynor can predict a somewhat unique portfolio in which a

Fig. 3. MPT results over the full period.

534
T.G.H. Surtee, I.P. Alagidede _
Borsa Istanbul Review 23-3 (2023) 527–540

Table 5
Example of unique Treynor predicted portfolios.
April–June 2017 July–Sept. 2017
Asset 1 wt Asset 2 wt Asset 3 wt returns Asset 1 wt Asset 2 wt Asset 3 wt returns
Sharpe 21.7% 9.8% 7.5% −7.6% 14.0% 10.8% 10.8% 5.4%
Sortino 21.1% 7.9% 6.7% −8.5% 13.2% 11.1% 10.3% 3.8%
Sterling 22.9% 8.7% 7.1% −8.5% 14.5% 11.0% 10.9% 4.0%
Treynor 9.8% 5.4% 4.5% −12.1% 6.7% 6.7% 6.7% 14.9%

loss could quickly be recouped. To show the principle, Table 5 we evaluate the similarity of the predictions and actual results
presents the weights of the three top assets selected by each in our 19 interactions. Using Excel's paired sample of means t-
model over the April–June 2017 period, when a significant loss test, we accept the null hypothesis for all four ratios, indicating
occurred. The market recovered in the next period (July–Sep- that the predictions and actual results could be significantly
tember 2017). The Treynor portfolio recovers the most in the different.4 The findings suggest that, theoretically, the MPT
next period. Longer-term studies could investigate whether perpetuity assumptions may not apply when portfolios are
Treynor predictions are good for recovery scenarios. rebalanced over shorter periods.
Treynor is different from the other three ratios in that it Further, we conduct two sets of paired one-tailed z-tests.
optimizes betas. Betas can give insight into why Treynor out- The first set is between the SASI and the six MPT models, and
performs 8 out of 19 times. Betas are calculated for all four the second set is between the Kagiso Fund and the six MPT
ratios. The data are grouped into betas, in which Treynor models.5 The null hypothesis is that the MPT models are the
should have been chosen (8 times) and when they should not same as the SASI and Kagiso Fund (control samples). Because
have been. Correlation and R-squared (coefficient of determi- all the p-values are less than our alpha, we reject all the null
nation) statistics are calculated to test whether the two groups hypotheses, meaning that our models outperform the control
differ. Table 6 tabulates the results. samples. Another interesting observation is that the Best
Although the data sample may be small, the table shows that Returns comparisons yields a zero p-value. It could be attrib-
Treynor produces the best portfolio when the other three ratios uted to computational rounding in Excel. However, it indicates
have a relatively strong negative correlation. This is observed that the results are highly significant. It appears including more
in the data, in which Treynor recovers from a previous loss, but Sterling and Treynor iterations in the best-performing model
the other ratios do not. There is a strong positive correlation has the most significant results, warranting further study of the
when Treynor is not the best solution. Similarly, R-squared Sterling and Treynor ratios.
shows that when Treynor is the preferred model, a small In line with the Literature, we calculate the kurtosis and
portion of the model's performance can be attributed to the skewness of the 19 iteration results. The comparison is between
benchmark SASI (where SASI represents the beta). The case the ratios in question (six Sterling or eight Treynor) and any of
for the 11 other iterations is different. The correlation and R- the other three ratios as a group—as in the Literature, the
squared suggest that when the Treynor model predicts a port- Sterling results, where leptokurtic, mostly have negatively
folio that is relatively divergent from the other three, it could skewed distributions.6
yield higher returns and vice versa when they are correlated. Table 7 shows the summary statistics for Sterling's kurtosis
This could be investigated further. and skewness. The average and standard deviation of the six
times Sterling is chosen as the best choice are compared to the
6.2. Predictiveness and distributions 13 times when the other ratios are chosen. Sterling is a good
choice if the average distribution is negatively skewed, but the
Thus far, we have discussed the best performance using the standard deviation's skewness might not be suggestive. The
four ratios. However, we did not look at their consistency in Treynor results differed from the Literature in that the platy-
terms of their predictions. MPT assumes portfolios with known kurtic distribution predicted portfolios with higher returns.
risk-reward ratios in perpetuity. We rebalance every three Table 8 presents the summary statistics showing that the
months. Hence, we do not know whether the suggested port- average kurtosis and skewness are negative, meaning that the
folio would have stabilized to the prediction over time. Instead, distribution has fatter tails with more data on the positive side
of the average. Similarly, the kurtosis and skewness of the
standard deviations are markedly lower when Treynor is the
Table 6
Summary statistics and unique predicted Treynor portfolios.
Sharpe Sortino Sterling Treynor 4
Where the p-values were less than our alpha (0.05).
Treynor as the best Correlation −0.44 −0.49 −0.53 1 5
The six MPT models include the four ratios' portfolios and the Subjective
solution (8 iterations) R-squared 0.20 0.24 0.28 1 and the Best Results portfolios. The Kagiso fund is the best performing sharia-
Other ratios as the Correlation 0.98 0.78 0.76 1 compliant fund. Other funds could be used, depending on investor subjectivity.
best solution R-squared 0.96 0.61 0.57 1 6
As in Table 4, when choosing the best-performing portfolio, Sterling would
(11 iterations) have been chosen six times.

535
T.G.H. Surtee, I.P. Alagidede _
Borsa Istanbul Review 23-3 (2023) 527–540

Table 7 Table 9
Sterling ratio distribution: summary of calculations. Examples of unique Treynor predicted portfolios.
Average Standard deviation Sharpe Sortino Sterling Treynor
Sterling Other ratios Sterling Other ratios Average Magnitude: MPT 18.46% 20.35% 22.27% 17.18%
Kurtosis 0.0031 −0.5857 0.0033 0.4805 Average Magnitude: Modified 29.85% 28.32% 29.73% 40.73%
Skewness −0.7040 0.0839 0.4371 0.3879 CAPM
Standard Deviation: MPT 11.11% 13.26% 13.15% 14.53%
Standard Deviation: Modified 16.42% 15.48% 19.29% 21.96%
CAPM
best choice. The negative skewness predicts a higher proba- Correlation between errors 0.35 0.27 0.01 −0.31
bility of higher-than-average returns in both cases. The current r^2 0.12 0.07 0.00 0.10
Literature suggests that only leptokurtic distributions are pre-
dictive, but we found evidence that platykurtic distributions
could also predict favorable portfolios. Further studies with predicted or time-lagged return estimations. To calculate the
bigger sample sizes could provide more precise suggestions. magnitude of errors, we take the absolute value of the error as a
The platykurtic, negatively skewed distribution should also be percentage of the actual return.
investigated for other ratios. In practice, investors who want to Table 9 presents the results of the MPT and CAPM
use this screening method could analyze the current data and magnitude of errors. MPT gives more consistent predictions
apply it one period ahead, as depicted in Fig. 2. when the average and standard deviation of the magnitude
between the 18 iterations is lower than that of the modified
6.3. Robustness testing CAPM. Next, we test for correlation between MPT and the
modified CAPM. The last two rows show that the degree of
We used the modified CAPM to test for robustness. The correlation and the R-squared determination between the two
model is reproduced here with the error (εit ) as the subject of methods are relatively low. Treynor is negatively correlated to
the equation: the CAPM. It stands to reason, as the CAPM uses betas, and
Treynor minimizes beta. We can conclude that although both
εit = − αi − βi0 (Rmt − Rft ) − β′i Zt−1 (Rmt − Rft ) + Rit − Rft methods (MPT and modified CAPM) have some errors, the
MPT models tend to be more accurate and more robust than the
where:
CAPM.
Rit is excess return on asset i in the period t (net of the risk-
As established in this study so far, Treynor could yield
free rate)
better results if it becomes more evident when to pick a
αi is the abnormal performance of asset i Treynor-based portfolio. First, we consider a paired sample for
(Rit −Rft ) is excess returns of the portfolio over a benchmark
a means t-test between MPT and CAPM errors in conducting
at time t
the analysis. However, because we rebalance the MPT every
Rmt are excess returns on the benchmark asset
three months, which affects the predictions into perpetuity, we
(Rmt −Rft ) is the excess return of the benchmark at time t
exclude the t-test analysis. Further research with different beta
βi0 is the systemic risk for the asset i for the current period frequencies may yield more insight and should be investigated.
(β′i Zt−1 ): β′i is the vector of conditional beta related to the
Economic significance is better for noting practical impor-
time-lagged predetermined instrument Zt−1 .
tance without being prescriptive (Mitton, 2021). Unlike other
εit is the forecast error for asset i at time t.
scientific undertakings where the order of magnitude is
The model uses historical data for each of the variables. β′i is
considered by factors of 10, we look at how growth and returns
time-lagged. Hence, we exclude the first iteration. For consis-
are reported. Industry practice reports growth as a percentage
tency, the historic betas used are the same covariate, weighted
of the initial investment. In our case, we questioned the sig-
betas derived in the MPT models. By making εit the subject of
nificance of our 3.5 times over performance of our best port-
the equation (see above), we effectively take the current per-
folio against the Kagiso fund. We considered our results
formance and subtract all other calculated performance factors.
economically significant. First, indexes and current funds do
The error is then the unexplained performance. We create a
not omit zakat when reporting their results. Hence, these
similar measure for MPT. Hence, we calculate the errors as the
benchmarks are overreported at face value. Omitting zakat
difference between actual and predicted returns. The errors
from the SASI decreases returns to 8.3 percent over five years,
could be negative if the actual returns are lower than the
whereas our worst-performing portfolio grows by 15 percent.
Our returns are 1.8 times the SASI and could subjectively be
deemed economically significant. Similarly, Kagiso out-
Table 8
Treynor ratio distribution: summary of calculations. performed the worst-performing fund by 1.9 times. Further-
more, the best-performing portfolio is a magnitude of 12 over
Average Standard deviation
the zakat-adjusted SASI and 3.5 times more than the Kagiso
Treynor Other ratios Treynor Other ratios
fund.
Kurtosis −0.8038 0.2261 0.6975 3.8901 Similarly, based on the results in Table 9, we deem the
Skewness −0.0431 0.1897 0.5023 0.8946
standard deviation's magnitude of errors between our results
536
T.G.H. Surtee, I.P. Alagidede _
Borsa Istanbul Review 23-3 (2023) 527–540

and the CAPM (between 1.4 and 2.3 times in favor of our frequencies. The algorithm is iterative and is illustrated in
results) economically significant. The test for robustness and Fig. 4.
economic significance of returns and errors shows the need for The various ways in which frequencies for the ratios can be
further research on portfolio optimization. calculated are shown in Appendix 2. These various ways of
using the data introduce increasing levels of algorithmic
6.4. Future research complexity and statistical inference. The complexity level in-
creases as different frequencies of all assets are calculated in
To the best of our knowledge, this is the first paper to use the first part of the algorithm, and covariant matrices created
the Sterling and Treynor ratios in MPT in this way. Hence, for each frequency in the second part, where the same fre-
throughout the discussion, we note avenues for further quency is used for the entire data set. Different frequencies can
research. To sum up, further research to test different frequency be calculated for both the numerator and denominator when the
data, consider less-constrained or unconstrained MPT models, compound annual rate of returns and the average largest
or derive different versions of the ratios will enrich knowledge drawdown (ALD) are calculated for the Sterling ratio, further
on this subject. increasing the complexity of the algorithm and the derived data
For example, Sharpe and Treynor use total volatility, size. At this stage, all the ratios for each asset can be compared,
whereas Sortino and Sterling use drawdown volatility. The use and the researcher can investigate optimal frequencies to use in
of drawdown beta volatility with the Treynor ratio could be the second part of the algorithm. Increasing the complexity
analyzed. The model becomes: might cause the optimal frequencies between the assets to
differ.
Rx − Rf
Drawdown Treynor Ratio = Furthermore, different ratios may be optimal at different
Drawdown βx frequencies, further expanding the data set. Iterative algorithms
can calculate the covariant matrices over all the data generated
where:
in the first part and apply them in the second part to determine
Rx is the expected return
the overall efficiency. Researchers will have to interrogate the
Rf is the risk-free rate (market proxy)
data and determine how to deal with any missing data problems
Drawdown βx is the drawdown betas of the expected return
between frequencies. After the matrices are derived, MPT will
Similarly, because Sterling is the next-best model, a hybrid
yield solutions, as in this initial study.
Sterling-Treynor ratio could provide further insight into port-
The frequency analysis and algorithms discussion show
folio optimization. The model would be:
several research disciplines for further study. Studies can
CARR improve portfolio predictions, optimization algorithms, and
Hybrid Ratio = ⃒⃒ ⃒,
covariance (Drawdown βx )⃒ statistical inference. MPT is used in fields beyond financial
analysis. There are opportunities for studying optimization
where: more broadly, with ratio interpretations not considered before.
CARR is the compound annualized rate of return. Policy implications are beyond the scope of this study. From
covariance (Drawdown βx ) is the covariate matrix of the perspective of sharia, different countries could have other
drawdown betas restrictions that can be introduced as constraints. Further, fund
Further, we note the limits of Excel as an optimization tool.7 managers may have policies that mandate asset weighting or
However, Excel is often used for research and discussed in the the number of assets in a portfolio. They can also be con-
Literature (see Google Scholar, 2022). We find Excel advan- strained in MPT algorithms.
tageous because the user interface and tabular layout offer re-
searchers a good visual platform for performing the 7. Conclusion
calculations and understanding how our new models differ
from prior models. The study is about improving performance, In an attempt to improve investment possibilities for Muslim
and our results support improvements with this novel approach. investors, we undertook to find high-return portfolios with
Further research focused on optimization with specialized minimal risk. Investors can pick portfolios in many ways, such
languages such as Python, R, or MatLab will be advantageous. as interpreting the news, reading financial statements, and
These languages can easily accommodate a more complex al- relying on their intuition. We argue that risk-return ratios such
gorithm that considers dynamic frequency periods. The algo- as the Sharpe and Sortino ratios, take the net effect of all in-
rithms would have two distinct parts. vestors’ positions on an asset. Modern portfolio theory (MPT),
The first part of the algorithm accommodates that different together with the Sharpe and Sortino ratios, is a common tool
assets are traded at different frequencies. Hence it calculates used to predict portfolios. However, we find that MPT could be
different frequency risk-reward ratios for each asset to deter- used with other risk-reward ratios, such as the Sterling or
mine their most favorable frequencies. The second part focuses Treynor ratios typically used for post-period evaluation.
on calculating covariances and optimizing the best rebalancing Standard Sterling and Treynor ratios are not conducive for
use in MPT; hence, we reinterpret them. We calculate periodic
compound rates of return for the Sterling ratio and calculated
7
The Excel model used and the parameters set are outlined in Appendix 1. beta as a covariance matrix for the Treynor ratio. Using these
537
T.G.H. Surtee, I.P. Alagidede _
Borsa Istanbul Review 23-3 (2023) 527–540

Fig. 4. MPT algorithm for optimizing different frequency risk-reward data.

calculations, we simulate five years (19 iterations) of portfolios studied using specialized optimization algorithms. Optimiza-
for all four ratios, selecting portfolios to maximize returns and tion algorithms of increasing complexity can be designed in
minimize risk. stages and tested to reveal statistical and economic signifi-
We conclusively find that our models could outperform the cance. Lastly, we posit that, although the context of our
current market indexes and funds and be more robust than the research is the sharia environment, we believe that research in a
CAPM, with economically significant results. However, further more general environment could be conducted.
analysis of the resulting distributions was needed to improve
our selection. Using kurtosis and skewness improved pre- Declaration of competing interest:
dictiveness. Kurtosis and skewness play a part in selecting
portfolios, and further study is warranted on how they can The authors declare that there is no conflict of interest.
support economic significance. Additionally, because redefined
post-period ratios proved to offer improved performance, we
suggest that further studies be conducted with different ver- Appendix 1.
sions of risk-reward ratios. Moreover, data frequency can be
calculated in many ways when these ratios are applied. These 1. The MPT model in Excel.
ratios can also offer significant improvements and should be

538
T.G.H. Surtee, I.P. Alagidede _
Borsa Istanbul Review 23-3 (2023) 527–540

Using the Sharpe ratio = ln(ptσ/px t−1 ) , for example, the opti- Contreras, O. E., Lizama, C., & Stein, R. A. (2016). The old ways are (some-
mization problem was set out in Excel using matrix multipli- times) the best: The performance of simple mean-variance portfolio optimi-
zation in various markets (pp. 1–25). Investment Company Fact Book.
cation (MMULT) equations. Portfolio returns are calculated as Davies, R. J., Kat, H. M., & Lu, S. (2016). Fund of hedge funds portfolio
the weighted excess returns ωi ln(pt /pt−1 ) for each asset. The selection: A multiple-objective approach. In S. Satchell (Ed.), Derivatives
assets and their weights form vectors that are multiplied to and Hedge funds (pp. 45–71). London: Palgrave Macmillan UK.
yield a scalar product. The result becomes the numerator of the DeFusco, R. A., McLeavey, D. W., Anson, M. J., Pinto, J. E., & Runkle, D. E.
Sharpe ratio. (2015). Quantitative investment analysis. Hoboken: John Wiley & Sons.
Elton, E. J., & Gruber, M. J. (1997). Modern portfolio theory, 1950 to date.
Next, the weights and covariance of returns of the assets are Journal of Banking & Finance, 21(11–12), 1743–1759.
multiplied to derive a scalar portfolio standard deviation (the Goetzmann, W., Ingersoll, J., Spiegel, M. I., & Welch, I. (2002). Sharpening
denominator). A variable is created to ensure that the weights Sharpe ratios. https://www.nber.org/papers/w9116.
of the assets add up to 100 percent. Similar calculations were Google Scholar. (2022). Excel as an optimization tool. https://scholar.google.
derived for the other ratios. com/scholar?hl=en&as_sdt=0%2C5&q=excel+as+an+optimization
+tool&btnG=/.
The Excel optimization model uses generalized reduced Hawley, J., & Lukomnik, J. (2018). The long and short of it: Are we asking the
gradient nonlinear optimization. We set the optimization right questions: Modern portfolio theory and time horizons. Seattle Uni-
objective to maximize the risk-return ratio by changing the versity Law Review, 41(2), 449–474.
weights. The optimization model weights are constrained to Hübner, G. (2005). The generalized Treynor ratio. Review of Finance, 9(3),
≥0 (zero) to eliminate short selling (negative weights). Each 415–435.
Kaplanski, G., Levy, H., Veld, C., & Veld-Merkoulova, Y. (2016). Past returns
asset was given equal weight as a starting point for the model. and the perceived Sharpe ratio. Journal of Economic Behavior & Organi-
The constraint precision was set at 0.001, and the number of zation, 123, 149–167.
iterations was set at 1000. The parameters were the same for all Kolbadi, P., & Ahmadinia, H. (2011). Examining Sharp, Sortino and Sterling
four ratios and all 19 iterations. All the iterations converged, ratios in portfolio management, evidence from Tehran stock exchange.
giving optimal portfolio suggestions. International Journal of Business and Management, 6(3), 222–236.
Magdon-Ismail, M., & Atiya, A. F. (2004). Maximum drawdown. Risk
2. Calculating ratios with different frequencies and Magazine, 17(10), 99–102.
increasing complexity. Ma, F., Li, Y., Liu, L., & Zhang, Y. (2018). Are low-frequency data really
The table shows how research can develop the optimization uninformative? A forecasting combination perspective. The North Amer-
algorithms of this research further. ican Journal of Economics and Finance, 44, 92–108.

MPT model variation Sharpe Sortino Derived Sterling Derived Treynor Other
derived ratios
Single frequency optimization for both numerator and X X X X X
denominator for all assets
Optimized ratio based on a single frequency taken after X X X X X
calculating a range of frequencies for both numerator
and denominator for all assets
Optimized ratio based on frequencies taken after X X X
calculating a range of different frequencies for
numerators and different frequencies for
denominators for all assets
Optimized ratio based on a single frequency taken after X X X
calculating a range of frequencies for both numerator
and denominator for each asset
Optimized ratio based on frequencies taken after X X X
calculating a range of different frequencies for
numerators and different frequencies for
denominators for each asset

References Marketing, J. S. E. (2013). Market data. https://web.jse.co.za/raise-capital/

equities-market/.
Markowitz, H. (1952). Portfolio selection. The Journal of Finance, 7(1), 77–91.
Ashraf, D. (2013). Performance evaluation of Islamic mutual funds relative to
Masih, M., Kamil, N. K., & Bacha, O. I. (2018). Issues in Islamic equities: A
conventional funds: Empirical evidence from Saudi Arabia. International
literature survey. Emerging Markets Finance and Trade, 54(1), 1–26.
Journal of Islamic and Middle Eastern Finance and Management, 6(2),
Mitton, T. (2021). Economic significance in corporate finance. SSRN 3667830.
105–121.
Pettengill, G. N., Sundaram, S., & Mathur, I. (1995). The conditional relation
Ashraf, D., & Mohammad, N. (2014). Matching perception with the reality:
between beta and returns. Journal of Financial and Quantitative Analysis,
Performance of Islamic equity investments. Pacific-Basin Finance Journal,
30(1), 101–116.
28, 175–189.
Rice, B. (2017). The upside of the downside of modern portfolio theory (pp.
Basilio, M. P., de Freitas, J. G., Kämpffe, M. G. F., & Rego, R. B. (2018).
13–18). Investment & Wealth Monitor, Investment Management Consul-
Investment portfolio formation via multicriteria decision aid: A Brazilian
tants Association.
stock market study. Journal of Modelling in Management, 13(2), 394–417.

539
T.G.H. Surtee, I.P. Alagidede _
Borsa Istanbul Review 23-3 (2023) 527–540

Rollinger, T. N., & Hoffman, S. T. (2013). Sortino: A “Sharper” ratio. Chi- Simo-Kengne, B. D., Ababio, K. A., Mba, J., & Koumba, U. (2018).
cago: Red Rock capital. http://www.redrockcapital.com/assets/RedRock_ Behavioral portfolio selection and optimization: An application to in-
Sortino_white_paper.pdf. ternational stocks. Financial Markets and Portfolio Management, 32(3),
Scholz, H. (2007). Refinements to the Sharpe ratio: Comparing alternatives for 311–328.
bear markets. Journal of Asset Management, 7(5), 347–357. Uppal, R., & Zaffaroni, P. (2016). Portfolio choice with model mis-
Schulmerich, M., Leporcher, Y.-M., & Eu, C.-H. (2015). Modern portfolio specification: A foundation for alpha and beta portfolios. SSRN https://ssrn.
theory and its problems. In M. Schulmerich, Y.-M. Leporcher, & C.-H. Eu com/abstract=2697866 https://doi.org/10.2139/ssrn.2697866.
(Eds.), Applied asset and risk management: A guide to modern portfolio Xiong, J. X., & Idzorek, T. M. (2011). The impact of skewness and fat tails on
management and behavior-driven markets (pp. 101–173). Springer. the asset allocation decision. Financial Analysts Journal, 67(2), 23–35.
Sharpe, W. F. (1994). The Sharpe ratio. Journal of Portfolio Management, Yahoo. (2021). Yahoo finance data. https://finance.yahoo.com.
21(1), 49–58.

540