Machine Earning – Algorithmic Trading Strategies

for Superior Growth, Outperformance and Competitive Advantage

Saïd Business School, University of Oxford, UK

Nicholas Burgess
[email protected]
[email protected]
29th March 2021

Did algorithmic trading generate superior returns relative to discretionary

trading during the Covid19 pandemic and do they provide a sustainable

competitive advantage?

Abstract

In this paper we use the tools and frameworks from Oxford University’s postgraduate

diploma in financial strategy to study the performance and benefits of algorithmic trading

strategies (algos), and specifically those that use artificial intelligence (AI) and machine

learning (ML).

We discover using valuation theory from (SBS2, 2020) that algos generate superior returns

compared to human discretionary trading both in normal market conditions and during

large market drawdowns, such as during the coronavirus (Covid19) pandemic. Furthermore

applying financial strategy techniques from (SBS1, 2020) we find that algos could be

Electronic copy available at: https://ssrn.com/abstract=3816567

combined with existing core competencies at my organization RUS 1 to create a sustainable

competitive advantage and give RUS an edge over its competitors.

Finally considering M&A growth strategies from (SBS4, 2020) we conclude that for RUS

algorithmic trading capabilities would be best acquired taking an organic approach as an in-

house build approach would be both cost-effective and allow for a more customized and

bespoke integration.

Even if only a fraction of the potential benefits are monetized, algo trading could have a

significant positive impact on earnings, which in turn would allow for reinvestment to

facilitate sustainable growth and maintain a sustainable competitive advantage.

1. Introduction

The current Covid19 pandemic required that the financial services industry invest heavily in

technology and cyber security in order to facilitate remote working, whilst adhering to strict

regulation controls to keep businesses and data secure. Opportunistic and prudent

investment firms could leverage such technology investments to invest in high growth

opportunities and increase their algorithmic trading capabilities.

In times of crisis trading systems using artificial intelligence & machine learning have the

potential to provide a competitive advantage as they constantly adapt to market conditions

(JP Morgan, 2019). They have the ability to process vast amounts of traditional, social media

and alternative reference data at high speed to gauge market sentiment. Moreover, they

1
See appendix for full case study details

Electronic copy available at: https://ssrn.com/abstract=3816567

can manage more highly diversified portfolios to reduce idiosyncratic risk and increase

Sharpe ratios2 (Esposito, 2020), (Institutional Investor, 2020).

In this paper we study if algorithmic trading strategies generate superior returns, if they

could provide a sustainable competitive advantage relative to discretionary (or human)

trading and we consider how to best acquire and integrate algo capabilities at RUS.

We proceed as follows: firstly we give an overview of the current market environment.

Secondly we introduce algorithmic trading and outline the main machine learning

techniques and how they can be used. Thirdly we compare the performance of algos and

discretionary trading systems using a hedge fund case study, where we classify funds by

their investment strategy. Fourthly we investigate if algos could provide a sustainable

competitive advantage and fifthly we consider how my organization RUS should acquire and

integrate algo capabilities with existing skills and resources.

2. Current Environment

The ongoing coronavirus (Covid19) global pandemic (2019-2021) has caused widespread

distress, disruption and as of January 2021 has claimed the lives of more than 2.2 million

people, (WHO, 2021). The attempts of Governments to control the virus through lockdowns

and curfews have harmed economies and businesses, particularly those in the hospitality,

travel and tourism sectors. Digital organisations, technology stocks and the FANGs3 have

2
Sharpe ratios measure performance as the return of an investment per unit risk.
3
The FANGs are the four prominent tech stocks in the U.S. namely Facebook, Amazon, Netflix and Alphabet
(previously Google). The equivalents in China are the BAT stocks Baidu, Alibaba and Tencent.

Electronic copy available at: https://ssrn.com/abstract=3816567

strongly outperformed (Wigglesworth 2020), whereas many businesses unable to adapt to

online working have suffered losses, faced closure and bankruptcy (Skeel, 2020).

Despite the development of new vaccines to counter the pandemic, many are still

wondering what a recovery could look like and businesses need to consider the possibility of

tail risks such as coronavirus mutation and further pandemics.

The new normal way of life relies on technology, with day to day business being conducted

online using technologies, such as Skype, Microsoft Teams and Zoom. Consequently to

survive many businesses had no choice but to invest in technology & infrastructure to

facilitate the high demand for online working and manage the associated cyber security

risks.

To more formally assess the current macro environment and highlight opportunities and

threats relevant to the financial services industry we perform a PESTEL analysis (Whittington

et al, 2020). The analysis is categorised into six environmental factors, namely Political,

Economic, Social, Technological, Ecological and Legal and is often used to aid macro

forecasts and scenario analyses.

In (figure 2.1) we build upon and extend the PESTEL analysis from (Burgess, 2020b) to

incorporate the most recent Covid19 impact assessments on the broad macro environment

see (McKinsey, n.d.) and (McKinsey, 2021).

Electronic copy available at: https://ssrn.com/abstract=3816567

Figure 2.1 PESTEL Analysis for Financial Services

Source: Macro Scenario Analysis of Financial Services (Burgess, 2020b)

Key highlights stem from the broad economic impact of the Covid19 pandemic, which has

required quantitative easing and government stimuli to counter the effect of mandatory

lockdowns, workforce disruption and business closures. In the U.S. alone, the economy is

estimated to incur net GDP losses ranging from $3.2 trillion (14.8%) to $4.8 trillion (23.0%)

over a 2-year period. (Walmsley et al, 2020).

Furthermore investment in new technologies facilitates a mobile, flexible workforce and

enables remote working to keep businesses open, but also presents opportunities to exploit

competitor weakness and capture market share due to the different levels of business

readiness and technological capacity in the industry. Prudent and opportunistic investment

firms could invest in technologies with a high growth potential, such as algorithmic trading,

to gain a sustainable competitive advantage.

Electronic copy available at: https://ssrn.com/abstract=3816567

3. What is algorithmic trading?

As outlined in (Burgess 2019a), algorithmic trading (or systematic trading) refers to the

automation of the trading process, through the creation of predefined rules (the ‘trading

system’) and their strict application when executing financial market transactions. The

automated trading system benefits from being repeatable and testable, unlike discretionary

trading (or human trading), which potentially has different rules for every transaction.

Algorithmic trading systems are predominantly employed by Hedge Funds to create

leveraged alternative investment portfolios uncorrelated with the market. Currently the

world’s hedge funds have USD 3.1 trillion assets under management (Statista1, n.d.).

Similarly algos are used by high frequency trading firms (HFTs) to make markets, seek

liquidity rebates and exploit market inefficiencies to benefit from arbitrage opportunities

(Brogaard et al, 2011).

Figure 3.1: Assets under management (AUM) of the largest hedge fund firms worldwide

in June 2020 (in USD billion)

Source: (Statista2, n.d.)

Electronic copy available at: https://ssrn.com/abstract=3816567

A recent survey by (BarclayHedge, 2018) asked managers of hedge funds and commodity

trading adviser funds (CTAs) for their insights and experience with AI and machine learning

and found that:

⮚ More than half managers use AI/ML to inform investment decisions

⮚ Two-thirds of funds use AI/ML to generate trading ideas and optimize portfolios

⮚ Over a quarter use automation to execute trades

⮚ Well over half have used AI for three or more years, and a third for five-plus years

Market drawdowns provide an excellent stress test of an algorithmic trading strategy’s

usefulness and ability to outperform. In 2020 financial markets were in turmoil and suffered

heavy losses due to the devastating impact of the coronavirus global pandemic. In the U.S.

markets plunged 38% in March 2020 to recover in April 2020 and the subsequent months.

This pattern was observed in many markets and exchanges globally.

Figure 3.2: U.S. Market Shock due to Covid19, S&P 500 (Mar-Apr 2020)

Source: Yahoo Finance

Electronic copy available at: https://ssrn.com/abstract=3816567

During the Covid19 pandemic hedge funds using algorithmic trading strategies suffered

heavy losses. Many trading strategies were trained to follow predefined trading signals that

were not able adapt to the new market environment. Consequently human-run hedge funds

trounced Quant funds (Bloomberg, 2020). For example in equity markets discretionary

funds beat systemic funds during the Covid19 market shock and performed much better

than Quants models and systematic investing (Factor Research, 2020).

Figure 3.3: Equity Market Neutral (EMN) Hedge Funds - Discretionary vs Systematic

Investing

Source: Factor Research

However artificial intelligence funds vastly outperformed discretionary funds with their

trading strategies having the ability to continually process, learn and adapt to new market

conditions. Reports such as (Institutional Investor, 2020) and (Robinson, 2020) claimed that

AI funds generated returns almost three-times higher than that of other hedge funds. Both

(Friedman, 2019) and (Eurekahedge, 2018) also confirm that AI and machine learning funds

generated superior returns as shown in (figure 3.4) and (figure 3.5) below.

Electronic copy available at: https://ssrn.com/abstract=3816567

Figure 3.4 Cumulative 3-Year Returns: AI Hedge Funds vs All Hedge Funds

Prequin performance benchmarks show that AI fund returns have outperformed the all hedge fund benchmark
by 3.09% with lower volatility and higher Sharpe ratios. AI funds reported 3.20% volatility and a Sharpe ratio of
1.96, while the hedge fund benchmark posted volatility of 3.87% and a Sharpe ratio of 1.40.

Source: Prequin Blog (Friedman, 2019)

Figure 3.5 Long-Term Analysis - AI vs Quants vs Traditional Hedge Fund Indices

AI hedge funds vastly outperform competing hedge funds, see (table 4.1) below for the fund definitions.

Source: Eurekahedge Database (Eurekahedge, 2018)

Electronic copy available at: https://ssrn.com/abstract=3816567

Consequently, from here onwards we focus on algorithmic trading systems that use artificial

intelligence, machine learning and predictive technologies to make investment decisions as

we look to answer the headline question and investigate if algorithmic trading generated

superior returns relative to discretionary trading during the Covid19 pandemic and if they

could provide a sustainable competitive advantage.

Machine learning funds try to incorporate and adapt to live market data and current

sentiment as much as possible. They go far beyond traditional back-testing approaches and

the use of fundamental and technical analysis to generate alpha. They look for patterns in

current market micro-structure4 and gauge current market sentiment, using behavioural

science, neuro-linguistic programming, natural language processing (NLP) and deep data

techniques to interpret and process enormous volumes of text, speech and sentiment from

social media, news channels and alternative reference data in real time, see (Lόpez de

Prado, 2018) and (Wyman, 2014).

The processing of vast amounts of data in real time is an impossible task for human traders.

Consequently discretionary traders are forced to concentrate their efforts on a handful of

securities compared to the machine learning funds that are able diversify far more broadly.

Algorithmic trading strategies characteristically hold a large number of securities to take

advantage of the law of large numbers and statistical edges. This diversification reduces

idiosyncratic risk, a risk that the Capital Asset Pricing Model (CAPM) says can be diversified

away and consequently investors are not rewarded for, see (Berk and DeMarzo, 2016),

(Brealey et al, 2014) and (Burgess, 2021). As a result machine learning portfolios have the

4
A simple example being the cancellation of many sell orders might indicate an imminent uptick or rally.

Electronic copy available at: https://ssrn.com/abstract=3816567

potential to be more efficient in the CAPM sense and to intrinsically carry lower risk, which

increases Sharpe ratios.

Machine learning techniques are broadly categorized as being supervised or unsupervised.

Supervised techniques are given pre-classified data to train the model, whereas

unsupervised learning techniques must discover trends, features, relationships and classify

data on their own. We outline several of the main machine learning classification

techniques below, see (Wilmott, 2019) for more information and examples of usage.

1. Kth Nearest Neighbours (KNN) – Supervised Learning

We start out with N data points that are already classified into groups or features. New

data is classified as being in the most popular group as determined by the Kth nearest

data points. For example, given a list of trading features that constitute a buy, sell or

hold signal, what does the current market features suggest we should do?

2. K Means Clustering (KMC) – Unsupervised Learning

Given unclassified data points represented by feature vectors5 we gather them into K

groups associated with their centres of mass or centroids. This is useful for measuring

financial data that clusters or is mean reverting, such as interest rate or volatility levels.

5
Each data point is represented as a list of features describing the data, commonly referred to as a feature vector

10

Electronic copy available at: https://ssrn.com/abstract=3816567

3. Naïve Bayes Classification (NBC) – Supervised Learning

Given samples of data representing different classes we calculate the probability of new

data being in each class. This technique is useful for natural language processing (NLP)

and analysing social media and news data for sentiment.

It uses Bayes Theorem, where P(A|B) = P(B|A).P(A) / P(B) to determine, for example, the

probability a news article with the words “Positive” and “Earnings” is a good news article

that will move stock prices upwards,

𝑃(𝐺𝑜𝑜𝑑 𝑁𝑒𝑤𝑠 | 𝑈𝑠𝑒𝑠 𝑡ℎ𝑒 𝑤𝑜𝑟𝑑𝑠 Positive, Earnings)

𝑃 (𝑈𝑠𝑒𝑠 𝑡ℎ𝑒 𝑤𝑜𝑟𝑑𝑠 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒, 𝐸𝑎𝑟𝑛𝑖𝑛𝑔𝑠 | 𝐺𝑜𝑜𝑑 𝑁𝑒𝑤𝑠 ). 𝑃(𝐺𝑜𝑜𝑑 𝑁𝑒𝑤𝑠)

=
𝑃(𝑈𝑠𝑒𝑠 𝑡ℎ𝑒 𝑤𝑜𝑟𝑑𝑠 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒, 𝐸𝑎𝑟𝑛𝑖𝑛𝑔𝑠)

4. Support Vector Machines (SVM) – Supervised Learning

Given a set of classified data represented by vectors of features, we divide the data

using a hyperplane and classify new data according to which side of the hyperplane each

data point lies. This is useful for detecting when trading levels are rich or cheap.

5. Self-Organizing Maps (SOM) – Unsupervised Learning

We start with all our data points represented as feature vectors. We then map similar

data and features into a two-dimensional grid to visualize which data have similar

characteristics. This is useful for hedging and identifying stocks for pairs trading. It is also

useful in Modern Portfolio Theory (MPT) whereby stocks can be grouped into buy and

sell categories based on expected returns, volatility and correlations between stocks.

11

Electronic copy available at: https://ssrn.com/abstract=3816567

6. Decision Trees – Supervised Learning

This is a flowchart technique and uses a hierarchy of features to divide (split) and classify

data. They are often referred to as classification and regression trees (CART). This could

be used to analyse winning and losing trades for example.

Figure 3.6 Decision Tree Example for Interest Rate Swap Transactions

The example tree estimates the probability of an interest rate swap transaction being profitable (winning)
given the transaction type exotic or vanilla, the currency of the swap and the maturity, where the decimal
inside the node represents the probability and the percentage of transactions in the category is outside
the node.

Source: Adapted from Wikipedia

7. Neural Networks – Supervised and Unsupervised Learning

A type of machine learning that tries to mimic brain activity. Signals are passed through

a network of neurons where they are mathematically manipulated and passed on for

further processing. They are used for modelling complex financial relationships,

establishing the functional form of data and pattern recognition.

12

Electronic copy available at: https://ssrn.com/abstract=3816567

8. Reinforcement Learning – Unsupervised Learning

This is one of the main types of machine learning, where the algorithm uses trial and

error and is based a system of rewards and penalties. The algorithm uses a heuristic

function to track the score of each trial run and is trained to maximize the score. This

method is useful for learning how to trade optimally in highly volatile markets and

interact with in environment when there is high uncertainty.

Trading strategies using supervised techniques often use historical data for pre-

classification, however past performance is no guarantee of future results. Unsupervised

learning approaches rely less on historical trends and seek to constantly learn and adapt to

new market environments. Consequently they require large representative data sets to be

effective, eliminate bias and error (Lynch 2018). Unsupervised strategies are perhaps well

placed to manage market shocks such as those faced during the Covid19 pandemic.

However it is a balancing act as fund managers don’t want trading systems to react too

quickly to noise in the market information nor so slowly that they miss a trend. Perhaps this

suggests that some human oversight may be helpful.

In summary algorithmic or systemic trading is simply the automation of the trading process,

which could involve full or partial automation of the trading system including trading signal

classification, trade execution, hedging and the risk management process. In theory such

trading systems are ideal to monitor, adapt and respond to new market conditions not seen

before, such as those experienced during the Covid19 pandemic, but how did they perform

in practice?

13

Electronic copy available at: https://ssrn.com/abstract=3816567

4. Algorithmic Trading Performance

In this section we study at algorithmic trading performance during normal market

conditions and market shocks such as that during the Covid19 pandemic (figure 3.2). We

perform a case study analysis comparing artificial intelligence tech fund performance with

that of quant and discretionary funds. We ask the reader to note that when evaluating and

comparing fund performance there are many factors to consider including investment

manager skill and ability, fund size, market sectors invested in, currency, inflation factors et

al, hence relative performance does not reflect on causality. Relative performance is purely

suggestive and used to highlight the potential for algorithmic trading and predictive

technologies to provide a competitive advantage.

4.1 Measuring Hedge Fund Performance

When comparing hedge fund performance it is not sufficient to only consider fund returns,

as investors can increase returns by leveraging positions and taking more risk. Consequently

fund performance is measured using risk-adjusted returns i.e. return per unit risk, which

gives a convenient uniform performance measure that can be applied to all funds. No

trading strategy or fund can be profitable at all times and during all market conditions,

consequently investors pay close attention to fund drawdowns, both the size and frequency

of losses suffered by the fund.

There are several measures of fund performance (Steinki and Mohammad, 2015), each with

different pros and cons. The main measure used is the Sharpe Ratio defined in (formula

4.1). Other measures include the Sortino Ratio that only measures downside risk (negative

variations) and the Treynor Ratio which measures risk as CAPM Beta (Burgess, 2021).

14

Electronic copy available at: https://ssrn.com/abstract=3816567

Formula 4.1 Sharpe Ratio (SR)

𝑟 −𝑟
𝑆ℎ𝑎𝑟𝑝𝑒 𝑅𝑎𝑡𝑖𝑜 =
𝜎
where 𝑟 denotes the return of a portfolio, 𝑟 the risk-free rate and 𝜎 the annualized

volatility or standard deviation of portfolio returns with all units in %.

Example 4.1 Sharpe Ratio

Consider two investment funds A and B having an average annual return 12% with

incremental drawdowns of 5% and 10% respectively. Clearly fund A performs best as it

bears less risk for the same return. If it is known the risk free rate is 2% for our investment

horizon then using (formula 4.1) we have Sharpe Ratio (fund A) = (12% - 2%) / 5% = 2.0 and

Sharpe Ratio (fund B) = (12% - 2%) / 10% = 1.0.

A fund’s performance is not only due to its ability to generate superior returns, but also its

ability to minimize risk and limit the size and frequency of its drawdowns. Consequently we

review fund performance not just in normal market conditions but also during the Covid19

market shock, as illustrated in (figure 3.2) when many funds were experiencing large

drawdowns. We investigate which funds generated superior returns on a risk-adjusted basis

and also examine which funds suffered fewer and smaller drawdowns during the Covid19

pandemic.

4.2 Hedge Fund Performance

To assess relative hedge fund performance we used Eurekahedge data. Eurekahedge is one

of the world’s largest hedge fund and private equity databases. For our analysis we used the

15

Electronic copy available at: https://ssrn.com/abstract=3816567

hedge fund indices and benchmarks outlined in (table 4.1). These hedge fund indices

represent collections of hedge funds by fund type.

Table 4.1 Eurekahedge - Hedge Fund Indices

These indices represent broad groups of hedge funds by fund type. The exact fund definitions are available
from (Eurekahedge n.d.) or via the Bloomberg terminal.

Source: (Eurekahedge n.d.)

Table 4.2 Eurekahedge - Hedge Fund Performance

Source: (Eurekahedge n.d.)

16

Electronic copy available at: https://ssrn.com/abstract=3816567

We extended a previous study (Eurekahedge, 2017) using (Eurekahedge n.d.) to collect and

collate more recent fund performance data covering the Covid19 period. We present a

summary of results above in in (table 4.2).

4.3 Algorithmic Trading Performance during Covid19 Drawdown (March-April 2020)

Hedge fund performance during market drawdowns is very important. It highlights the skill

of the fund manager to avoid large losses and manage downside risk. Summarizing the

results presented in (table 4.2) we see that during the Covid19 market drawdown in March-

April 2020 (figure 3.2) AI funds performed well and made gains of +3.27% whereas

traditional discretionary funds reported losses of -2.23% over the same period.

AI funds had the lowest downside volatility of 2.25% and amongst all the hedge fund groups

had one of the smallest maximum drawdowns, which for AI funds was -7.24% compared to

-6.30% & -14.42% for quant funds and -12.27% for discretionary funds.

Morningstar Direct data reported that during the Covid19 pandemic AI technology funds

held up relatively well, reporting that in April 2020 as the FTSE 100 and S&P 500 plunged

more than -20% and tech funds were down as little as -4.6% (Esposito, 2020). Bloomberg

also showed that the downside volatility of AI funds during the Covid19 major market

downturn was exceptionally small (figure 4.3).

17

Electronic copy available at: https://ssrn.com/abstract=3816567

Figure 4.3 Hedge Fund Comparative Returns

During the Covid19 market downturn and rebound in Mar-Apr 2020 (figure 3.2) the Eurekahedge Artificial
Intelligence Hedge Fund Index did not suffer large drawdowns and showed little volatility and downside risk
relative to the competing fund groups from (table 4.1).

Source: Bloomberg Terminal, COMP <GO>

Hedge funds using artificial intelligence consistently have the lowest drawdowns and best

risk-adjusted returns (Sharpe ratios), not just during the Covid19 pandemic, but more

generally in crisis situations and during major market risk events as shown in (table 4.4).

18

Electronic copy available at: https://ssrn.com/abstract=3816567

Table 4.4 AI / Machine Learning Hedge Fund Returns During Key Market Risk Events

Source: Adapted from (Eurakahedge 2017) to include Covid19 risk assessment.

This answers the first part of our headline question: “Did algorithmic trading generate

superior returns relative to discretionary trading during the Covid19 pandemic?” - Clearly

funds using artificial intelligence outperformed discretionary funds during the pandemic as

shown in (table 4.2) with gains of +3.37% and losses of -2.23% respectively.

4.4 Algorithmic Trading Performance after Covid19 Drawdown (Post April 2020)

Hedge fund performance after the Covid19 drawdown highlights how funds performed in

more normal and less extreme market environments. This highlights the fund manager’s

ability to profit in normal market conditions and more specifically in this period to also

benefit from and exploit post Covid19 opportunities.

Reviewing the results presented in (table 4.2) from a post Covid19 perspective we see that

algorithmic trading using artificial intelligence earned returns of +11.24% beating quant

funds with returns of +7.85% and +10.92% and almost matching traditional discretionary

19

Electronic copy available at: https://ssrn.com/abstract=3816567

funds with +12.26%. On a risk-adjusted basis AI funds outperform all funds and have the

highest Sharpe ratios6.

Equivalently, from a CAPM perspective, when plotting the risk-return profile for each hedge

fund group we see AI funds generate the highest returns per unit risk (figure 4.5). The

gradient of the lines in red through the fund returns and risk-free rate indicate the Sharpe

ratio, which for AI funds is the closest to the Capital Market Line (CML) indicating AI funds

are the most efficient on a risk-adjusted basis (Berk and DeMarzo, 2016), (Burgess, 2021).

Figure 4.5: Shape Ratios and Risk-Return Profiles for Hedge Fund Indices

For each of the funds from (table 4.1) we plot the annualized return and standard deviation. The slope of the
red-lines denotes the Sharpe ratio, quoted in blue. We implied the risk-free rate as 2.0% using (formula 4.1).
The black and blue lines are for illustrative purposes and represent the CAPM Capital Market Line (CML) and
Efficient Frontier respectively.

Source: Eurekahedge data from (table 4.2)

6
AI funds also have the highest Sortino ratios, where risk is measured as downside volatility.

20

Electronic copy available at: https://ssrn.com/abstract=3816567

In August 2020, (Institutional Investor, 2020) also reported that AI Tech Funds had vastly

outperformed other funds, having produced cumulative returns of 34 percent in the three

years through to May 2020, compared with a 12 percent gain for the global hedge fund

industry over the same period. Furthermore Bloomberg confirm that not only did AI funds

outperform in the post Covid19 environment, they have also outperformed on a long-term

basis as shown in (figure 4.6).

Figure 4.6 Long-Term Artificial Intelligence Hedge Fund Performance

Comparing the hedge fund benchmarks from (table 4.1) we see that artificial intelligence hedge funds vastly
outperform competitor funds.

Source: Bloomberg COMP <GO>

Having confirmed that algorithmic trading has the potential to generate superior returns

relative to discretionary trading we proceed to examine if they could provide a sustainable

competitive advantage.

21

Electronic copy available at: https://ssrn.com/abstract=3816567

5. Competitive Advantage

As outlined in (Burgess, 2020c) the long-term survival of any organisation relies upon its

ability to well-manage resources to create dynamic capabilities valued by customers.

Threshold capabilities are required to survive in a given market and achieve competitive

parity. However dynamic capabilities utilise the entire value chain (Zenger, 2013). They are

sets of resources, capabilities, skills and abilities (figure 5.1), which together combine to

form core competencies that can respond dynamically to environmental opportunities and

threats. They have the potential to create a sustainable competitive advantage.

Figure 5.1 View of the Value Chain

Source: SSRN Strategic Analysis of Japanese Megabanks (Burgess, 2020c)

22

Electronic copy available at: https://ssrn.com/abstract=3816567

Prior to assessing if algorithmic trading skills and capabilities could complement the existing

value chain we performed a SWOT7 analysis (Whittington et al, 2020), see (figure 5.2). This

was to assess RUS’s current capabilities, internal strengths and weaknesses and current

ability to manage the external opportunities and threats presented in (figure 2.1).

Figure 5.2 SWOT Analysis for RUS

Source: Strategic Analysis of Japanese Megabanks (Burgess, 2020c)

RUS are currently unable to monetize core competencies due to capital constraints, low

trading volumes and high cost to revenue ratios. Furthermore limited performance metrics

act as a business tax that disables management from understanding RUS’s value

7
An acronym for strengths, weaknesses, opportunities and threats

23

Electronic copy available at: https://ssrn.com/abstract=3816567

proposition, its core strengths and weaknesses. It also diminishes their ability to maximize

profits, reduce costs and manage key risks.

Furthermore using a competitor SWOT analysis (figure 5.3) to contrast RUS’s core

competencies against key competitors reveals that RUS are only able to achieve competitive

parity. RUS are better placed to manage external threats but poorly placed to exploit

external opportunities. The SWOT analysis suggests RUS are more risk averse than its

competitors. It is well placed to manage external threats from coronavirus workforce

disruption to regulatory Libor reforms (Burgess, 2019a). However it is poorly placed to

exploit advances in technology and lucrative government green finance initiatives.

Figure 5.3 Competitor SWOT Analysis

Source: Strategic Analysis of Japanese Megabanks (Burgess, 2020c)

24

Electronic copy available at: https://ssrn.com/abstract=3816567

A VRINO8 analysis (Galpin, 2020) helps to evaluate if, how and to what extent an

organisation has a value chain (figure 5.1) with resources and capabilities that when

combined can achieve and sustain a competitive advantage (Whittington et al, 2020).

We performed a VRINO analysis, in (Burgess, 2020c), based on RUS’s current capabilities

and value chain. In this paper we look extend this analysis to examine if investing in

algorithmic trading capabilities 9 and combining this with the existing “agile pricing and risk

analytics” (Burgess, 2020c) could create a new core competency “Advanced Automation of

Pricing, Risk and Execution” that could offer superior trading and risk management

capabilities and give RUS a sustainable competitive advantage. The VRINO analysis is

performed below.

5.1 VRINO Analysis - Valuable

The new core competency would facilitate sophisticated state-of-art trade execution and

orders could be worked systematically to reduce market frictions to achieve the best price.

This is valuable to customers as it would facilitate better transaction prices and reduce

human error.

It is valuable to RUS as it would also reduce human resource costs, which are the biggest

costs the organization incur. It would also reduce operational costs, create economies of

scale, improve cross-selling opportunities and would enable RUS to outperform

8
VRINO is an acronym for Valuable, Rate, Inimitable, Non-substitutable and Organisationally Appropriable,
sometimes also referred to as VRIO without the Non-substitutable element.
9
In this paper we are specifically referring to algorithmic trading systems that use artificial intelligence,
machine learning and predictive technologies.

25

Electronic copy available at: https://ssrn.com/abstract=3816567

competitors, charge clients lower fees, compete more aggressively on price to win market

share and provide elevated levels of client service.

There is a strong possibility algorithmic trading plays a key role in generating superior

returns. As demonstrated by hedge fund managers using artificial intelligence, who have

better managed market drawdowns and outperformed other funds on a risk-adjusted basis

as indicated by their higher Sharpe ratios (figure 4.5).

5.2 VRINO Analysis - Rare

We performed a strategy canvas analysis to produce a value curve for RUS and its

competitors (figure 5.4). A value curve measures perceived performance against critical

success factors (CSFs). It can be a subjective process and difficult for managers to agree on

which critical success factors to prioritise as highlighted by (Kim and Mauborgne, 2002). The

analysis highlighted that algorithmic trading capabilities, digital services and technology

innovation are rare skills that could exploit competitor weakness. Algorithmic trading

presents a lucrative, monopolistic, ‘Blue Ocean’ investment opportunity, where profit

margins are wider as markets are less competitive and less congested.

26

Electronic copy available at: https://ssrn.com/abstract=3816567

Figure 5.4 Strategy Canvas – Competitor Value Curve Analysis

Source: Strategy Canvas Analysis & Competitor Benchmarking, (Burgess, 2020e)

5.3 VRINO Analysis - Inimitable

It is extremely difficult to imitate the new core competency, because it is nuanced, complex

and requires tacit knowledge of the intricate connectivity between value chain components.

Furthermore it is unique to the bank making it non-transferrable and inimitable.

5.4 VRINO Analysis - Non-Substitutable

It would be expensive and complex to replace or substitute the new core competency. As it

would be costly to set-up a substitute and require niche technical expertise. Human

substitutes would be unable to able to process vast amounts of data, compete with the

economies of scale and the speed of performance. Similarly a technology or platform

27

Electronic copy available at: https://ssrn.com/abstract=3816567

substitute would require niche skills to build in-house. It would be equally difficult to

purchase and integrate external vendor solutions, as already experienced with Murex,

where the technology is expensive, nuanced, complex to configure and support due to the

niche skills and experience required.

5.5 VRINO Analysis - Organizationally Appropriable

The new core competency is highly organizationally appropriable and could help monetize

existing capabilities and facilitate many new revenue generating opportunities. Likely

benefits include:

Enhanced Returns

⮚ Increased investment returns and Sharpe ratios, see (table 4.2) and (figure 4.5)

⮚ Advanced Market Forecasting, Prediction and Trading Signal Capabilities

⮚ Increased Transaction Speeds and Ability to Exploit Arbitrage Opportunities

⮚ Automated Execution, Hedging and Advanced Order Book Management

Advanced Risk Management

⮚ Advanced Risk & Drawdown Management, see (table 4.2) and (table 4.4)

⮚ Reduction in Human Emotion/Bias, Human Error and “fat finger” mistakes.

⮚ Advanced Order Management - Can simultaneously Execute and Hedge

⮚ More Diversification Opportunities and Reduced Idiosyncratic Risk (Burgess 2021)

28

Electronic copy available at: https://ssrn.com/abstract=3816567

Improved Client Services & Market Share

⮚ Lower Transaction Costs & More Competitive Pricing

⮚ Ability to Work Large Orders without Moving the Market

⮚ Hybrid or Fully-Automated ‘Robo Research’ and ‘Robo Sales’ Services

⮚ Improved Client Services: Automation of Repetitive Tasks would Free-up Human

Resources for High Value Client Service Items

⮚ Advanced Research leveraging Broad Market, Social Media & Client Data Analysis

⮚ Improved Market Share and Client Coverage Capacity

⮚ Enhanced Cross Selling Opportunities

Cost Savings

⮚ Automation of Repetitive Tasks & Human Resource Savings

⮚ Lower Operational Costs

⮚ Scalable Services & Economies of Scale

Even if RUS could only monetize a fraction of these opportunities the new core competency

has the potential to increase revenues significantly, reduce risk and lower costs.

Furthermore RUS could opportunistically go after the quick and easy revenue enhancing

targets first and use the profits for sustained reinvestment and innovation.

The VRINO analysis indicates that investing in algorithmic trading would give a sustainable

competitive advantage as summarized in (figure 5.5).

29

Electronic copy available at: https://ssrn.com/abstract=3816567

Figure 5.5 VRINO Analysis

Source: Adapted from Strategic Analysis of Japanese Megabanks (Burgess, 2020c)

This answers the second part of our headline question: “do they provide a sustainable

competitive advantage?” - Yes they certainly provide a sustainable competitive advantage.

Specifically for RUS, they would enable existing resources and capabilities to be

organizationally appropriable, monetizable and enable RUS to improve its strategic position

from competitive parity to that of having a competitive advantage by shoring up

technological weaknesses to exploit market opportunities.

6. What should my organization do?

Having established that algorithmic trading technologies have high revenue generating

potential and could provide a sustainable competitive advantage for RUS (figure 5.5); how

should RUS acquire this skill and integrate it with existing resources and capabilities?

30

Electronic copy available at: https://ssrn.com/abstract=3816567

The goal of a Merger & acquisition is to acquire new resources and capabilities to build up a

firm’s existing capabilities, increase firm value and gain market share. Therefore we use

M&A tools and frameworks from (SBS4, 2020) and (Galpin, 2020), as they are perfectly

designed to answer this question.

In order to acquire algorithmic trading capabilities, RUS need to consider if it should build or

buy the resources and technologies required. We outline the pros and cons of each

approach in (figure 6.1).

Figure 6.1 Build vs Buy Strategies

Source: Winning at the Acquisition Game (Galpin, 2020)

RUS could pursue organic growth via in-house expertise and internal alliances. Alternatively

the core competency could be acquired via vendor software purchases, fintech

partnerships, joint-ventures or even a fintech acquisition.

31

Electronic copy available at: https://ssrn.com/abstract=3816567

An organic approach is most suitable for RUS and most compatible with RUS’s requirements

and concerns:

Cost Cutting & Capital Constraints

⮚ RUS is cost cutting, has capital constraints & looking for a cost effective solution

⮚ Need for niche technical expertise & careful cost management (Burgess, 2019a)

⮚ Make use of free open source vendor solutions (Scikit-Learn, n.d.) and (TensorFlow, n.d.)

Risk Aversion

⮚ Want to take an experimental approach before committing capital

⮚ Increased revenue opportunities must be demonstrable and evidenced

⮚ Want an incremental programme of work that focuses initially on low cost easy wins

⮚ Gradual change to minimize disruption

⮚ Workforce require time to train and adapt to new processes

Data Protection

⮚ RUS is concerned about proprietary data leakage & GDPR10

⮚ Vendors might share RUS alpha generation ideas to win new customers

⮚ A vendor non-disclosure agreement (NDA) is not considered sufficient protection

As illustrated in (figure 6.1) organic growth would allow RUS to take a cost-effective, risk-

averse and incremental integration approach. Organic solutions are tailored to business

10
The General Data Protection Regulation (GDPR) is a legal framework in the European Union (EU) that sets
strict guidelines for the collection and processing of personal information.

32

Electronic copy available at: https://ssrn.com/abstract=3816567

needs, highly configurable and over time they better help establish in-house expertise.

Downside challenges would likely arise from having to work with expensive legacy

infrastructure, outdated IT policies and poor IT management. Manager skill is critical for

success to navigate IT challenges and if managed well this approach could be a cost effective

solution.

Following (Ansoff, 1957) we recommend the integration plan illustrated in (figure 6.2). We

have high confidence in this approach as it was previously used successfully to establish the

“agile pricing and risk analytics” core competency.

Figure 6.2 Ansoff’s Growth Matrix

Source: Strategies for Diversification (Ansoff, 1957)

33

Electronic copy available at: https://ssrn.com/abstract=3816567

Ansoff Integration Plan

1. Market Penetration

First consolidate core business lines and focus on market penetration and growing core

rates trading and fixed income businesses, which are well-established yet have more

potential.

2. Product & Services Diversification

Second diversify by product and asset class to provide advanced services and support for

credit derivatives, foreign exchange, interest rate swaptions, convertible bonds and equity

businesses.

3. Market Diversification

Thirdly diversify by market, starting with the most similar business entities, namely RUS

Bank London, RUS Securities New York, Hong Kong then Tokyo (in this order). A cautious

more arm’s length approach should be taken with overseas branches, where cultural CAGE11

distances (Ghemawat and Altman, 2019) can lead to lengthy P&L12, cost, ownership and

other internal disputes that often destroy rather than create value.

When integrating algo capabilities, RUS should focus on enhancing core business

performance and limited related diversification. Research shows (Whittington et al, 2020)

that it is important to diversify, but not over-diversify, as unrelated diversification is

ineffective and often lowers performance, as illustrated in (figure 6.3).

11
CAGE distances track cultural, administrative, geographical and economic differences.
12
P&L is an acronym for profit and loss.

34

Electronic copy available at: https://ssrn.com/abstract=3816567

Figure 6.3 Diversification and Performance

Source: Exploring Strategy, (Whittington et al, 2020) pp 244-245

It will be important to establish internal alliances between quant and research teams to

steer the innovation, train the trading and sales teams and assist with data analyses to be

used for identifying cross selling opportunities and trading signals for alpha generation.

For RUS a cost-effective way to transition towards using machine learning techniques would

be to consider using free open source Scikit-learn and TenorFlow13 analytics to complement

the existing in-house analytics with PCA14 and other highly sought after machine learning

techniques, tools and frameworks for pricing and risk management.

Initially machine learning processes should be kept simple to use. To enhance revenue

opportunities machine learning must be implemented into processes in a way that is

13
Scikit-learn and TensorFlow are an open source analytics libraries that provide machine learning tools for
data analysis, see (Scikit-Learn, n.d.) and (TensorFlow, n.d.) respectively.
14
RUS is keen to explore the use of Principal Component Analysis (PCA), which is a technique for reducing the
dimensionality of a dataset to increase data interpretability whilst minimizing information loss.

35

Electronic copy available at: https://ssrn.com/abstract=3816567

complementary and organic to existing workflows in order to combine the best human

financial expertise with the best AI/ML techniques. This is often referred to as a

Quantamental15 approach (Lόpez de Prado, 2018), where we combine quantitative

approaches using high-end computers, mathematical models and big data with fundamental

methods where humans manually analyse investment opportunities to generate better risk-

adjusted returns (Lynch, 2018), (Smigel, n.d.). Moreover, as automation frees human

resource, human effort should be reassigned to high value-add work items.

RUS should incrementally enhance existing trading and sales capabilities to target earnings

growth16, starting by enhancing data analysis and the automation of well-understood tasks

and processes such as the below,

Targeted Algo Enhancements for Earnings Growth

1. Enhance Trader Analytics for Better Pricing, Risk and Trading Signals

2. Automate Research & Trading Ideas

3. Enhance Client Data Gathering & RFQ17 Analysis

4. Enhance Sales Processes to Identify Cross-Selling Opportunities

5. Auto Hedging (Humans Execute Main Deal, but with Auto-Hedging)

6. Auto Execution (with Human Oversight Only)

7. Advanced Execution Services to Work Larger Orders Efficiently

15
This is a portmanteau combining “quant”itative and fund”amental” investing
16
Increased earnings from both revenue growth and cost reduction
17
An electronic price request is called a Request for Quote (RFQ)

36

Electronic copy available at: https://ssrn.com/abstract=3816567

Figure 6.4 Projected Revenue Growth (%)

Source: Illustrative RUS Revenue Projections

Even if only a fraction of the benefits from (section 5.5) and (figure 6.4) are monetized,

algorithmic trading would have a significant positive impact on earnings. This would

facilitate reinvestment to develop the new core competency to maturity, establish further

economies of scale and facilitate sustainable growth.

7. Conclusion

We have learned that algorithmic trading strategies generated superior returns relative to

discretionary trading during the Covid19 pandemic and that they facilitate better

management of major market drawdowns. We also showed that both during normal market

conditions and times of crisis they generate superior risk-adjusted returns and exhibit higher

Sharpe ratios.

37

Electronic copy available at: https://ssrn.com/abstract=3816567

An algorithmic trading strategy would allow RUS to dynamically respond to market events,

better identify market opportunities, reduce costs, reduce operational risk, improve client

services, increase market share and establish economies of scale. It would complement

RUS’s existing value chain to create a new core competency “Advanced Automation of

Pricing, Risk and Execution” services, which would give RUS a sustainable competitive

advantage.

These are unprecedented times for the world and financial markets; the economic outlook

is uncertain yet cautiously optimistic. Coronavirus vaccines, continued workforce disruption,

weak economies and government financial stimuli with tax hikes to follow dominate news

headlines. There will be a clear paradox to both cut costs and innovate. We believe these

risks will be best managed through an organic algo growth strategy, which will allow RUS to

be innovative and stable (McGrath, 2012), whilst establishing a sustainable competitive

advantage.

Appendix Case Study

Organization: Rainbow & Unicorn Securities, London (RUS)

Industry: Financial Services

Sector: UK Corporate Lending & Project Finance

RUS provide financial services, trading expertise and risk management solutions within the

interest rate, fixed income and credit markets. It specialises amongst other things in

38

Electronic copy available at: https://ssrn.com/abstract=3816567

corporate lending and project finance within the UK and also Europe, Middle-East and Africa

(EMEA).

RUS is fully owned by Rainbow & Unicorn Financial Group Tokyo (RUFG), which has a sizable

balance sheet, predominantly due to large Japanese deposits. Japanese investors are

extremely risk-averse with many business relationships based on trust, reputation and long-

term track formed over decades!

RUFG is one of three ‘Japanese Megabanks’ that dominate the financial services industry in

Japan with combined deposits exceeding USD 10 trillion (Piece-of-Japan, n.d.). RUFG itself

has assets of USD 20 billion and employs 50,000 staff with 500 offices worldwide.

Figure A1: RUFG Organization Chart

RUFG has 3 regional hubs to provide concentrated coverage in Europe & EMEA, Asia &

Japan and the Americas. RUFG subsidiaries are siloed with Corporate Banking (B) ring-

39

Electronic copy available at: https://ssrn.com/abstract=3816567

fenced and Investment & Securities (S) businesses separated to satisfy legal and regulatory

requirements.

Corporate Banking subsidiaries (B) have large balance sheets, good access to capital but

poor legacy systems & infrastructure. Investment & Securities (S) businesses however have

good technology and systems, but small balance sheets and little access to capital.

This case study is fictional, yet inspired by adapted from real-world industry practice and

first-hand experience.

References

Ansoff I. (1957), Strategies for Diversification, Harvard Business Review, 1957, 35(5), pp. 113-124.

BarclayHedge (2018), BarclayHedge Survey: Majority of Hedge Fund Pros Use AI/Machine Learning in
Investment Strategies Available at: https://www.barclayhedge.com/insider/majority-of-hedge-fund-pros-use-
ai-machine-learning-in-investment-strategies

Berk J. and DeMarzo P. (2016), Textbook: Corporate Finance, Global Edition, 2016. Published by Pearson, ISBN
978-1-292-16016-0

Bloomberg (2020), Human-Run Hedge Funds Trounce Quants in Covid Year

Available at: https://www.bloomberg.com/news/articles/2020-12-30/human-run-hedge-funds-trounce-
quants-in-year-defined-by-pandemic

Brealey R. et al (2014), Textbook: Principles of Corporate Finance, Global Edition, 2014. Published by McGraw
Hill Education, ISBN 978-0-077-15156-0

Brogaard J. et al (2011), UK Government Review: The Future of Computer Trading in Financial Markets.
Available at:
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/289021/
11-1241-dr10-high-frequency-trading-information-and-profits.pdf

Burgess N. (2019a), An Introduction to Algorithmic Trading: Opportunities & Challenges within the Systematic
Trading Industry. Available at SSRN: https://ssrn.com/abstract=3466213

Burgess N. (2019b), Libor Benchmark Reform: An Overview of Libor Changes and its Impact on Yield Curves,
Pricing and Risk. Available at SSRN: https://ssrn.com/abstract=3479833

40

Electronic copy available at: https://ssrn.com/abstract=3816567

Burgess N. (2020a), Strategic Analysis of Japanese Megabanks – Analysis 1 of 5
Changes in Relative Prioritization of Stakeholders & the Implication for Corporate Strategy
Available at SSRN: https://ssrn.com/abstract=3645549

Burgess N. (2020b), Strategic Analysis of Japanese Megabanks – Analysis 2 of 5

Macro Scenario Analysis of Financial Services
Available at SSRN: https://ssrn.com/abstract=3645551

Burgess N. (2020c), Strategic Analysis of Japanese Megabanks – Analysis 3 of 5

Core Competency Development for Competitive Advantage
Available at SSRN: https://ssrn.com/abstract=3645553

Burgess N. (2020d), Strategic Analysis of Japanese Megabanks – Analysis 4 of 5

Financial Services Industry Analysis
Available at SSRN: https://ssrn.com/abstract=3660674

Burgess N. (2020e), Strategic Analysis of Japanese Megabanks – Analysis 5 of 5

Strategy Canvas Analysis & Competitor Benchmarking
Available at SSRN: https://ssrn.com/abstract=3660679

Burgess N. (2021), How Risky is your Project Really? Corporate Finance Strategies for Assessing Risk. Available
at SSRN: https://ssrn.com/abstract=3748822

Esposito A. (2020), Morning Star, How Tech Funds Performed in the Covid-19 Crisis, 7 April 2020. Available at:
https://www.morningstar.co.uk/uk/news/201203/how-tech-funds-performed-in-the-covid-19-crisis.aspx

Eurekahedge (n.d.), Eurekahedge is an independent data provider and alternative research firm that
specializes in hedge fund databases. Available at: https://www.eurekahedge.com/

Eurekahedge (2017), Eurekahedge Research: Artificial Intelligence: The New Frontier for Hedge Funds (1/2).
Available at: https://www.eurekahedge.com/Research/News/1614/Artificial-Intelligence-AI-Hedge-Fund-
Index-Strategy-Profile

Eurekahedge (2018), Eurekahedge Research: Artificial Intelligence: The New Frontier for Hedge Funds (2/2).
Available at: https://www.eurekahedge.com/Research/News/1724/Artificial-Intelligence-Hedge-Fund-Index-
Strategy-Profile-2-February-2018

Factor Research (2020), Hedge Fund Battle: Discretionary vs Systematic Investing.

Available at: https://www.factorresearch.com/research-hedge-fund-battle-discretionary-vs-systematic-
investing

Friedman B. (2019), Prequin Blog: The Rise of the Machines: AI Funds Are Outperforming the Hedge Fund
Benchmark. Available at: https://www.preqin.com/insights/research/blogs/the-rise-of-the-machines-ai-funds-
are-outperforming-the-hedge-fund-benchmark

Galpin T. (2020), Textbook: Winning at the Acquisition Game – Tools, Templates and Best Practices Across the
M&A Process. Oxford University Press, ISBN 978-0-19-885856-0

41

Electronic copy available at: https://ssrn.com/abstract=3816567

Ghemawat P. and Altman S. (2019), The State of Globalization in 2019, and What it Means for Strategists.
Harvard Business Review, Feb 2019

Institutional Investor (2020), AI-Powered Hedge Funds Vastly Outperformed, Research Shows. Available at:
https://www.institutionalinvestor.com/article/b1mssrswn1mpr0/AI-Powered-Hedge-Funds-Vastly-
Outperformed-Research-Shows

JP Morgan (2019), JP Morgan Asset Management Portfolio Insights, Machine learning in hedge fund investing -
Alternatives for uncorrelated income. Available at: https://am.jpmorgan.com/ca/en/asset-
management/institutional/insights/portfolio-insights/machine-learning-in-hedge-fund-investing/

Kim W. and Mauborgne R. (2002), Charting Your Company’s Future. Harvard Business Review, June 2002.

Lόpez de Prado M. (2018), Textbook: Advances in Financial Machine Learning

John Wiley & Sons, 2018 (1ed), ISBN: 978-1-1109-48208-6

Lynch V. (2018), TechCrunch Publication: 3 ways to avoid bias in machine learning

Available at: https://techcrunch.com/2018/11/06/3-ways-to-avoid-bias-in-machine-learning

McGrath R. (2012), How the Growth Outliers Do It. Harvard Business Review, Jan-Feb 2012

McKinsey (n.d.), McKinsey & Company: Impact on economies - Data graphics, insights from leaders, and stories
that illustrate the impact of COVID-19 on economies. Available at: https://www.mckinsey.com/about-
us/covid-response-center/covid-19-impact/impact-on-economies#Macroeconomic

McKinsey (2021), McKinsey & Company Survey: The coronavirus effect on global economic sentiment, Jan
2021. Available at: https://www.mckinsey.com/business-functions/strategy-and-corporate-finance/our-
insights/the-coronavirus-effect-on-global-economic-sentiment

Robinson G. (2020), International Investment Case Study: Covid-19 strengthens the case for AI in fund
management – study. Available at: https://www.internationalinvestment.net/news/4018595/covid-19-
strengthens-case-ai-fund-management-study

SBS1 (2020), Course in “Financial Strategy” as part of Postgraduate Diploma in Financial Strategy taught by
Richard Whittington, Säid Business School, University of Oxford.

SBS2 (2020), Course in “Business Finance” as part of Postgraduate Diploma in Financial Strategy taught by Ilaria
Piatti, Säid Business School, University of Oxford.

SBS3 (2020), Course in “Corporate Valuation” as part of Postgraduate Diploma in Financial Strategy taught by
Martin Schmalz, Säid Business School, University of Oxford.

SBS4 (2020), Course in “Financial Strategy for Growth” as part of Postgraduate Diploma in Financial Strategy
taught by Timothy Galpin, Säid Business School, University of Oxford.

Scikit-Learn (n.d.), Scikit-learn User Guide: Simple and Efficient Tools for Data Analysis. Available at:
https://scikit-learn.org/stable/user_guide.html

42

Electronic copy available at: https://ssrn.com/abstract=3816567

Skeel D. (2020), Report: Bankruptcy and the coronavirus, 21 April 2020. Available at:
https://www.brookings.edu/research/bankruptcy-and-the-coronavirus/

Smigel L. (n.d.), Analyzing Alpha Blog: “Quantamental: What is it & Why it works”.
Available at: https://analyzingalpha.com/quantamental

Statista1 (n.d.), Value of assets managed by hedge funds worldwide.

Available at: https://www.statista.com/statistics/271771/assets-of-the-hedge-funds-worldwide/

Statista2 (n.d.), Assets under management of the largest hedge fund firms worldwide in June 2020. Available
at: https://www.statista.com/statistics/273133/assets-under-management-of-the-largest-hedge-fund-firms/

Steinki O. and Mohammad Z. (2015), Common Metrics for Performance Evaluation: Overview of Popular
Performance Measurement Ratios. Available at:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2662054

TensorFlow (n.d.), TensorFlow: An end-to-end open source machine learning platform Available at:
https://www.tensorflow.org/

Walmsley T. et al (2020), Economics of Disasters and Climate Change Journal: The Impacts of Coronavirus on
the Economy of the United States. Available at: https://doi.org/10.1007/s41885-020-00080-1

Whittington R. et al (2020), Textbook: Exploring Strategy 12th edition, Pearson Publishing 2020, ISBN 978-1-
292-28251-0

WHO (2021), World Health Organization Coronavirus Disease (COVID-19) Dashboard

Available at: https://covid19.who.int/

Wigglesworth R. (2020), The financial Times – How Big Tech got even bigger in the Covid-19 era, Robin
Wigglesworth in Oslo, 1 May 2020. Available at:
https://www.ft.com/content/d2e09235-b28e-438d-9b55-0e6bab7ac8ec

Wilmott P. (2019), Textbook: Machine Learning – An Applied Mathematics Introduction

Panda Ohana Publishing 2019, ISBN 978-1-9160816-0-4

Wyman O. (2014), Marsh & McLennan Companies Publication: The Hidden Alpha in Equity Trading – Steps to
Increasing Returns with the Advanced use of Information. March 2014. Available at:
https://www.oliverwyman.com/content/dam/oliver-
wyman/v2/publications/2014/mar/The_Hidden_Alpha_in_Equity_Trading.pdf

Zenger T. (2013), Value Chains & Systems, Value Creation Maps and Strategy Metrics to Know the Value of
your Firm. What Is the Theory of Your Firm? Harvard Business Review, June 2013.

43

Electronic copy available at: https://ssrn.com/abstract=3816567