Rethinking Asset Allocation Part 2

The COVID-19 crisis has sped up several structural trends and triggered an urgent need to rethink deeply how investors approach financial markets. Indeed, the quickest we learn how to adapt to the increased “evolutionary pressure”, the fastest we will overcome and succeed in the aftermath of the global pandemic. In this sense, the extent by which AI and Machine learning empower us to thrive in this uncharted territory is clear: they are very powerful tools to improve our investment decision-making, and consequently, build better and more resilient portfolios. 

In the second issue of this two-part paper, we share new insights to help asset and investment managers face these unprecedented times. From reconsidering how classical theories remain a valuable compass for investors during turbulent times, to weighing the benefits of AI to build a better portfolio positioning, these insights will help frame where opportunities lie, and where the latest breakthroughs in the field of machine learning and quantitative finance are leading the world of investing. How will asset allocation look like in the 2020s?

‍

Introduction 

Whether we like it or not, the world we used to live in has changed in the aftermath of the global pandemic. Yet, although it is early to tell what will be the long-term consequences, it is clear that we experienced a rapid – almost overnight – shift that calls investors to rethink how they approach financial markets, portfolio construction and risk management.

Indeed, from the sharp unbalances in oil supply and demand occurred in late March, to the massive government and central bank intervention, financial markets have not remained immune. Rather, they seem to have taken the bullet of a deep (and unexpected) economic crisis that has very few comparisons in recent history. 

In this sense, if it is true that the pandemic has sped up several structural trends, it is also true that technology, especially Artificial Intelligence, has turned out to be vital to deal, survive and adapt to this new complex world. As it becomes a mature technology, it is also rapidly becoming our most powerful ally to improve our ability to make better decisions based on a scientific and rational approach to financial markets. In this sense, it marks the crucial shift from human and statistical reasoning to a new era of assisted decision-making.

In the first half of this two-part paper, we retraced the footsteps of the academic and empirical research surrounding asset allocation, pointing at the direction towards which the industry is moving. We examined how AI models are enhancing the analysis of asset correlations and investment factors, together with the benefits associated with dynamic portfolio rebalancing strategies and ESG investing.

In this second part, we will explore why classic economic theories at the heart of asset management are still a valuable compass for investors. We will also examine the extent by which Artificial Intelligence (especially deep neural networks) is allowing us to go beyond the traditional approach to portfolio optimization to build more stable portfolios. In this sense, we will discuss the reason why AI is a game changer for traditional investment research and how this translates into not only a better understanding of financial markets and portfolio positioning, but also in how it is helping scientists to build more robust and insightful financial models.

Start integrating AI into your investment process with the most advanced no-code solution.

Fill out the form to book a demo of Sphere®, and learn how it can bring value to your investment decisions.

Book A Custom Demo

By submitting you are agreeing to the Terms of Use and Privacy Policy
Thank you! Your submission has been received.
Oops! Something went wrong while submitting the form.

Why Modern Portfolio Theory is Still Relevant

Similarly to how travelers, sailors or soldiers use maps to orient themselves, investors are used to rely on financial theories as their compass to navigate and understand financial markets’ dynamics. However, just as travelers do not expect their maps to be 100% correct, investors do not expect their models to be perfect, but only to capture the essential elements. Put differently, maps and economic theories become useful guidance for the information they omit, because they filter out the noise and focus only on what is relevant for making decisions.

‍

We can generally trace a line between two different approaches in building a map, or in this case economic theories. On the one hand, there are the so-called “descriptive” theories that use empirical and deductive reasoning to depict how something behaves in a specific situation. On the other hand, “normative” theories attempt to frame the essential features and prescribe an optimal behavior under a set of initial assumptions (e.g. Modern Portfolio Theory).

‍

In this sense, the work of the economist and Nobel-prize winner Harry Markowitz in the 1950s has set the stone for what we call today the “modern approach to investing”. The framework he laid down for portfolio construction established for the first time a repeatable – and efficient – way for investors to build efficient portfolios and evaluate their asset allocation decisions using the information provided by three key elements: the level of risk tolerance, the expected return on each security, and their correlations among each other. 

‍

Being a normative theory, Markowitz’s theory was originally (and intentionally) developed considering a very simplified investment setting. Indeed, it described the process that a rational investor would have made in an ideal world – a world where transaction costs are absent, there are no informational asymmetries and in which every investor invests at the same time in the most efficient portfolio available on the market. 

‍

However, although the world he described differed from the realistic representation of how financial markets behave, under this set of assumptions it was easier to show what it took to create efficient and well-diversified portfolios: select assets with negative or low correlations, so to smooth out individual gains and losses.

‍

Eventually, Markowitz was the first capable of drawing an effective model to build portfolios by applying the scientific method. In this sense, the restrictive assumptions of the model have not reduced its capacity to serve as useful guidance for investors. Nowadays, the core principle, and improved versions of the original model, are still widely used. Building diversified and efficient portfolios has remained the top priority of any asset and investment manager looking to obtain the best returns for any given level of risk tolerance. 

‍

Under this light, we understand the extent to which normative theories serve as very powerful tools to face the complexity of financial markets. Yet, rather than being a crystal ball, Modern Portfolio Theory represents a one-of-a-kind “theoretical ecosystem”, which has set the foundations of efficient asset allocation but that can nonetheless be continuously expanded and improved. Indeed, over time researchers have gradually dropped the original simplifying assumptions to make the model more coherent to the underlying market dynamics (e.g. transaction costs), to address specific problems initially not accounted for (e.g. short-selling, multiperiodal setting) or to introduce new tools that perform better than traditional ones in some specific areas (e.g. deep neural networks applied to the analysis of the variance-covariance matrix).

‍

Eventually, what appears to be extremely insightful is that, due to the increasing complexity of financial markets, the adoption of Artificial Intelligence into this framework is quickly becoming an essential element for its future development. Thanks to its unprecedented power and precision, AI is able to better extract the signal in the noise of the markets, and adapt to the gradual unfolding of financial markets.  In this sense, like a needle that always points North, Markowitz’s ideas are today more relevant than ever. AI is just taking them to the next level: from a compass to a fully-fledged GPS.

‍

Maths Holds the Secret Ingredient of Performance

Between the 19th of February and the 23rd of March, the S&P 500 Index fell by almost 34%, the largest drawdown since the 2008 financial crisis. On the 16th of March, in just a single day, the index sank by almost 12%. This global equity selloff that followed the spread of COVID-19 across the globe has not only scared investors, experienced and novice ones alike, but also reminded them of some fundamental lessons in finance. 

‍

The first lesson is how big of a role diversification plays in asset allocation. In this sense, a portfolio only focused on US equity would have faced far greater losses compared to a diversified portfolio of global equities, bonds, commodities and alternatives. The second consequential insight is something well known in the industry but rarely emphasized enough: in terms of capital appreciation, controlling volatility is incredibly beneficial in the long run. Indeed, reducing fluctuations is being seen as an increasingly fundamental element to improve financial returns. The reason behind this lies in the fact that extreme fluctuations destroy portfolio performance. 

‍

To better grasp this concept, let’s imagine having $1,000 to invest on the 18th of February. If we had invested everything on the S&P 500 Index, in just a little over a month we would have experienced a catastrophic -34% loss – leaving us with just $660. Even if we had posted an equal gain of 34% the following month, we would have reached not our initial $1,000 but just $885 – an 11,5% drop.

‍

On the contrary, if we had invested in a hypothetical better-diversified portfolio that had fallen by just 10% and then gained the same amount the following month, our losses would have been significantly lower. In fact, after a month we would have $900 and, after two, $990 – just a 1% decline.

‍

The table of Exhibit 1 expands our previous example, showing the behavior of four other hypothetical portfolios with different levels of diversification and thus volatility. Each portfolio has an initial negative return, followed by a positive return of the same percentual amount. Despite the arithmetic mean being always 0%, the portfolios end up with significantly different total returns – that is, with different final performances.

‍

Exhibit 1 – Portfolios with different levels of diversification and different levels of volatility. 

‍

From the table above we can get an immediate sense of why controlling portfolio volatility is essential to improve their performances. Indeed, when a portfolio experiences a large negative drawdown, it is significantly harder to get back to the original $1000. For instance, after portfolio D went down by -50%, an investor would need to achieve a +100% return to breakeven. On the contrary, a portfolio with lower volatility would require a not-so-extreme positive return to recover its loss - in the case of portfolio B, a +5,26% return after sinking -5%.5

Indeed, the benefit of a less volatile portfolio allows managers to smooth out negative drawdowns and accelerate positive gains. This behavior stems out of what is known mathematically as the compounding effect – something that the Nobel-prize winner Albert Einstein once called “mankind’s greatest invention”.

Compounding can be indeed very powerful if it plays by our side. As we calculate the performance of our investments accounting for the previously accumulated return, maintaining over time smoother fluctuations translates into capital growing at a much faster rate over the long term. At the same time, it can also be detrimental if we experience significant volatility that destroys performance. From this standpoint, we can think of compounding as a double-edged sword that must be handled with extreme care. For investors, understanding how to limit its negative downside and avoiding sharp rises and drawdowns is indeed of the utmost importance. 

Reducing portfolio volatility is thus incredibly important, and the best way to do so is by efficiently diversifying it. As we have discussed in the first issue of this two-part paper, the evolution of Artificial Intelligence and Machine Learning is enabling investors to gain a deeper understanding of correlation dynamics and thus achieve better diversification.

From this standpoint, great diversification does not only lead to better performance because the best or the right investment is chosen. Instead, it does so because – by mitigating drawdowns – investors protect the capital on which returns are calculated. Consequently, staying invested in the markets with an effective diversification strategy enables asset and portfolio managers to substantially accelerate their capital appreciation process and safely build wealth over time.

‍

Watch Out: Optimization is Powerful—Not Magical

Portfolio optimization is one of the most powerful tools we have to manage investments. Just as we look at fuel economy when we plan on buying a new car, we are used to looking at the Sharpe – the reward-to-risk ratio – of our investments to see if we are allocating capital efficiently or if we take too much risk compared to return. Yet, achieving efficiency takes much more effort than investing in the portfolio that had the best Sharpe ratio in the past. As we are going to see, the reason for this is twofold: the efficient portfolio is not stable over time and optimization processes can often be flawed. 

As we discussed in a recent paper entitled “Sharper than Sharpe”, maximising forward – rather than past – efficiency is what really matters for investors, as it comes from having understood and correctly interpreted the underlying dynamics of financial markets and it is the result of better investment decisions.

Despite the value that investors now pose in portfolio optimization, judging if a portfolio was optimal or not was fairly discretional before the quantitative approach to investing began to be widespread. Indeed, it was not until the development of the above-mentioned Modern Portfolio Theory that a wide consensus was reached. 

As we have seen in the first chapter, this theory gave asset and investment managers a mathematical, repeatable framework to identify optimal portfolios by analyzing just three variables: risk, return and correlations. Through a mathematical technique called constrained optimization, investors found the so-called “efficient frontier” – the set of portfolios that maximized the tradeoff between return and risk. From that standpoint, they could choose the best portfolio according to their own risk profiles.

In this sense, improving a portfolio’s efficiency – the ratio between return and risk – has become a primary goal for a majority of investors, using the Sharpe Ratio as their preferred indicator. This metric compares the expected return in excess of the risk-free rate with the standard deviation of a security over a given time interval.

After its first introduction, there have been many refinements to portfolio optimization. Yet, this technique, if not handled with care, can still mislead investors because of two main reasons: the efficient frontier varies as time goes by to accommodate the mutated correlation dynamics and the optimization process is only as good as its input data.

First and foremost, financial markets are increasingly dynamic. Consequently, the efficient frontier is not static but continuously changes over time. Indeed, asset managers must always look at their investment decisions not as point-in-time but rather as multiperiodal. Every decision they make – even the ones considered optimal at a certain time – must be updated periodically to reflect the evolution of market conditions. 

From this standpoint, Exhibit 2 considers the case of how the risk profile associated with a 60/40 portfolio (i.e. a portfolio invested 60% in fixed income and 40% in equity) would have changed over about 60 years of financial markets, from the 1960s to 2020. As we can see, not only the efficient frontiers move sharply from one decade to the other, but the same asset allocation has completely different risk profiles.

‍

Exhibit 2 — Efficient frontier and the 60/40 portfolio from 1960 to 2020. Equity and Fixed Income are proxied by the S&P 500 Index and the Bloomberg Barclays Aggregate Bond Index respectively

‍

Secondly, the optimization process is a powerful tool but not a magical one. If low quality, incorrect or biased data are used, then several estimation errors may occur. Indeed, investors can fall prey to several biases, both in the methodology they use and in the sample they decide to analyze. Among these, one of the most common and dangerous ones is the risk of overfitting – creating over-complex and case-specific models that fail to generalize the actual relationship and thus to provide actual insights.

‍

Indeed, optimality does not automatically turn into a well-performing portfolio. As a matter of fact, efficiency can be regarded as the effect, rather than the cause, of performance. For this reason, investors are increasingly adopting forward-looking and dynamic optimization frameworks to closely follow the gradual evolution of financial markets to maximize future – not past – efficiency. In this sense, transitioning from a traditional to a forward–looking approach to portfolio optimization is how we can build portfolios that are stable, efficient and well-positioned to face the evolution of financial markets.

‍

Portfolio Positioning? Hope for the Best, but Prepare for the Worst

Financial markets are an intrinsically turbulent environment, as they tend to react quickly (and sometimes excessively) as soon as new information becomes available. Because of this, financial data is often characterized by a very low “signal-to-noise ratio” – meaning that on top of the huge quantity of data available, very little actually conveys the information we need to make informed decisions. 

‍

In this sense, the recent global pandemic crisis has highlighted once again how difficult it is for asset and investment managers to effectively manage risk and deploy successful strategies in an increasingly unpredictable world.

‍

Investors are becoming increasingly aware of the need to better protect their portfolios from unexpected events and shocks. In the words of economists John Kay and Mervyn King’s famous book, “Radical Uncertainty: Decision-making for an unknowable future”, the best way for investors to stay resilient and flexible in an unpredictable world is to account for what they call “alternative future scenarios”.

‍

Traditional tools, such as stress testing and scenario planning, have been utilized with this purpose for a long time. Indeed, simulating possible market shifts and analyzing the portfolio behavior at both a granular (i.e. its individual securities) and holistic (i.e. the overall portfolio) level represents one of the most common practices in the industry. Yet, traditional quantitative techniques seem to be struggling to adapt to the constant evolution of financial markets.

‍

Indeed, until a few years ago, researchers faced a big computational obstacle when performing this kind of task. As each simulation needed to be properly set up and analyzed, analysts were often unable to explore efficiently all possible future combinations. As a consequence, they were often forced to make assumptions and estimations that necessarily reduced the potential benefit of stress testing. To a certain extent, they shifted from the risk of not doing stress tests, to the risk of not doing it properly. 

‍

In this sense, the progress achieved in the fields of Artificial Intelligence and advanced statistical analysis is gradually bridging this gap, allowing investors to build efficient and diversified portfolios capable of adapting to the constant evolution of market conditions. In other words, they are enhancing what is called portfolio positioning – the ability of a portfolio to efficiently endure the gradual unfolding of financial markets by discounting the possible evolution in asset correlations.

 

Indeed, asset allocation choices need to factor in not only historical data and the current market scenario, but also taking in consideration their expected evolution in a range of forward-looking scenarios. This allows for a twofold benefit: on the one hand, it allows to paint a more complete picture of what we can expect from the portfolio and, on the other hand, it contributes to diversifying actively from being overly exposed to a given source of risk.

‍

From this standpoint, the growth and the adoption of AI in finance are enabling asset and portfolio managers to perform hundreds of thousands of market simulations, bringing stress testing to a whole new level. A probabilistic – rather than static – forecasting approach is leading investors to a 360-degree understanding of the potential future shifts in market dynamics. In this sense, asset and portfolio managers are learning how to dynamically adapt to the gradual unfolding of the markets, mitigating volatility and positioning their portfolios so that, if turbulent periods arrive, they avoid putting all their investments at risk.

‍

Complex, not Complicated. Is AI the Financial Vaccine we Needed?

Only a few years ago, it was rare to find applications of Artificial Intelligence in the world of finance. Yet, after its successful applications in areas such as credit ratings, trade execution and anomaly detection, also the asset management industry has rapidly moved towards its adoption, to the point that today it is hard to find an area where AI is not being deployed in some form and does not add value to the investment process. 

‍

Indeed, as larger data sets, greater computing power, and more efficient algorithms become available to investors, Artificial Intelligence has quickly grown into a mature technology that can assist investors to improve their investment decision-making, and consequently, build better and more resilient portfolios. 

‍

In particular, one area which has gained substantial traction from the introduction of AI has been the development of more robust financial theories and models. This happens because AI can play the double (and key) role of decoupling the search for variables from the search for the model’s specification. Indeed, considering the complexity of financial markets, it is unlikely that a researcher will be able to uncover a new financial theory by a visual inspection of the data or by running a few regressions. 

‍

Indeed, a solid theory allows us to support our decisions and build robust investment strategies. Most importantly, it allows us to see the bigger picture instead of just the tip of the iceberg, highlighting the causal relationship between the different pieces of the economic puzzle. In this sense, the reason why AI is proving particularly successful in this area is due to the fact that it has also transformed how investment research is performed, substantially adding a twofold benefit. 

‍

On the one hand, it has allowed scientists to dramatically reduce bias during their research activities. For example, machine learning models (e.g. deep neural networks) do not require to be specified ex-ante but shape themselves as they learn from data by uncovering hidden patterns. This poses an upper limit to how much scientists can (sometimes unconsciously) alter the model’s output, moving towards an unbiased discovery of the relevant relationship in data.

‍

On the other hand, researchers have also leveraged AI to automate and bring to scale a large part of their repetitive tasks, such as analyzing vast amounts of data sequentially. Indeed, it has allowed scientists to focus on more important and valuable tasks such as supervising the learning process of the AI model and examining far more evidence, rather than raw data.

‍

In this sense, this points at a critical observation: the need for theoretical understanding far outweighs the agnostic search for better predictions. This represents the reason why the most insightful use of Artificial Intelligence in finance is for discovering new theories, and also why the traditional statistical and econometric toolbox is not equipped to deal with it - because financial markets are complex (i.e nonlinear), not complicated (i.e. linear). Rather than replacing theories, AI plays the vital role of helping scientists form theories based on rich empirical evidence. Eventually, AI is just a tool – just an incredibly powerful, fast and intelligent one. By leveraging its superior data analysis power, AI is enabling investors to meet two goals at the same time: extract more signal out of the input data and build flexible models that adapt over time to new market conditions.

‍

Exhibit 3 — How traditional and AI–driven models look at market dynamics

‍

The Road Ahead

Especially during a period of unprecedented change, innovation cannot stop. Rather, innovation is nearly foundational to navigate out of the crisis itself. This dramatic global pandemic has only reinforced this concept in our minds, something we all already knew but were less conscious about. Technology is any investors' best ally. Being at the forefront of it means to have the best possibility to reap its benefits.

This two-part paper has explored how AI is changing the world of asset allocation and what are the key insights that we can learn from it. All investors entering the new decade should consider whether they are using the right tools to navigate this complexity efficiently. 

From taking investment research to the next level to helping build more robust financial theories and portfolios, the gains that AI is bringing to the field of portfolio management are indeed pervasive. In this sense, these mature tools have certainly helped investors efficiently navigate today’s extraordinary times – and are poised to lead tomorrow’s “new normal” ones.

Start Integrating AI Into Your Investment Process