SHARE

Absa Alternative Asset Management: Developing a framework to track the progression of COVID-19

We set out to give clarity on several key questions around the emergence of the COVID-19 virus pandemic; what is the prognosis for the disease in terms of global spread, can the current measures being taken arrest the spread and is South Africa taking the right steps to fight the pandemic?  Additionally, we examine the current impact of the spread on the global economy and define a framework that gives a probabilistic measure in tracking the chances of the US falling into a recession as a result of the impact. While the analysis gives context to the current situation, the evolution of the crisis will continue to play out – Absa Alternative Asset Management (AAM) believes that the defined tools will allow for an ongoing pragmatic picture of the developing situation.

The emergence of the COVID-19 pandemic in January 2020 will have a significant impact on both social and economic aspects of the global environment for the foreseeable future. To understand how the pandemic may play out, it is critical to get an understanding of the various driving factors. As a predominately quantitative asset manager, AAM takes an initial look at several aspects relating to the developing pandemic and the resulting potential impact on asset prices. Two broad factors are presented; a brief overview on the mathematical modelling of pandemics and the current global response, as well as a machine learning analysis of the current macroeconomic top-down environment. Clearly, the situation is unfolding along an unprecedented path with disruption and potential crisis that has not been comparatively experienced since the Second World War.

 

Basic mathematics of pandemics

Attempts to understand the driving factors behind the dynamics of epidemics date back to 1917[1]. The authors describe several factors relating to both the nature of the pathogen and external elements that may aggravate the severity of an outbreak. In 1927, the SIR model was developed by Kermack and McKendrick[2], a stochastic, or time-based formulation used to model the progression of an outbreak through time.

Fundamentally, the model considers a population of susceptible individuals (S) into which a pathogen, causing infection (I), is introduced. Individuals who fall ill either then recover (R) through time or die (the model only considers case closure for these individuals when they no longer represent a threat in making others ill, regardless of the outcome). The severity and resultant shape of the outbreak depends on two specific inherent characteristics of the pathogen itself: the ease with which it is able to infect the population (known as the infection or attack rate) and its mortality rate. A high mortality rate with low infection rates tend to burn out quickly. Assuming a population of 1 000 has been introduced to one sick individual with a disease, which has a high infection rate and low mortality expectation, the SIR predicts the outcome as shown in Figure 1 below.

Figure 1: Simple depiction of the SIR formulation for the mathematical modelling of infectious diseases [2, 9, AAM].

There is a rapid rise in number of infections (represented by the orange line) through time as healthy individuals (blue line) succumb to the infection. The number of active cases peaks at about 675 as the number of cases close either via recovery (green line) or death. Once the number of infections reach the maximum level, cases begin to decline steadily. Clearly, this is a very simplistic view of the population and as such assumes that the entire cohort becomes infected rapidly. This scenario would be the case if the entire global population is susceptible to the virus with all individuals falling ill and ultimately finding an outcome one way or the other. In reality, rapid increases in infected individuals lead to a severe overload of the global health care system, as is being experienced currently, and efforts to flatten this curve assist significantly. The curve may be flattened due to inherent features of the virus, such as high and swift mortality or slow infection rates. Social distancing, as well as early tracking of cases, testing of suspected infections and isolating cases, all assist in flattening the active infection peak and ensuring that the majority of the population do not succumb to infection. Figure 2 below shows the progression of this through time.

Figure 2: Simple depiction of the SIR showing the effect of reduced infection rates[2, 9, AAM].

Although 80% of this cohort becomes infected, this happens in a much slower and less severe fashion, and in reality allows the health systems to manage the impact of severe cases. Moving towards this eventuality in a global sense is critical, with efforts at social distancing now required to ensure that the COVID-19 virus does not follow a path of rapid infection. Initial work done by various epidemiologists suggests that the COVID-19 virus is relatively containable[3, 4] if the responses by governments and individuals are appropriate.

The modelling of these types of infections has become much more sophisticated and beyond the scope of this report. However, this basic understanding does allow for a better comparative view in analysing the current global pandemic situation. AAM analysed the development of the current pandemic in line with the presented theory above. All data was sourced from the Johns Hopkins School of Public Health [5]. All countries’ current data (as of 22 March 2020) was constructed to show the active infection cohort (orange line in Figures 1 and 2 above), as well as the recovered/deceased profile, representing all cases with an outcome. All analyses compare the development trajectory since the confirmed or active cases reached 200 casualties and compare the progression on a day-count basis and as such is not reflective of a specific date but rather days into the infection. Figure 3 below shows the profile for China.

Figure 3: China’s COVID-19 pandemic, showing infected cases and outcomes [5, AAM].

As shown, the active case load had significant initial growth – the large jump on day 21 is as a result of the inclusion of additional tests related to patients’ lung scans. In light of the typical profile expected by the SIR model, China seems to have contained the initial outbreak – current cases are around 5 770, with just over 75 000 individuals no longer infectious, with the total confirmed cases at 81 397. China was extremely proactive in limiting the spread of the disease and locked down the Hubei province when the confirmed number of cases reached 761.

Through their actions, China managed to peak out the active infection rate after 26 days, which was predicted by Chinese epidemiologists [3]. South Korea also managed to limit their outbreak and curbed the infection peak in just over 20 days. The Chinese experience now gives a comparative template with which to compare other countries’ trajectories. Figure 4 below plots the largest exposures of European countries, as well as the United States.

Figure 4: Comparative active infection rates of countries with elevated attack rates [5, AAM].

Most countries exhibit a typical infection path, as defined by the exponential growth rates associated with infection dynamics and SIR models. China appears to follow a far more linear relationship. One potential explanation for this is that the infection had been in the environment for some time before testing had begun. The combined recovery/death curves, however, are in line with expectations and are tracked closely by the comparative countries – most notably, Italy, Spain and Switzerland. Germany and the US seem to have very slow outcome curves*.  Alarmingly, the US exhibits much greater infection rates than any of the comparative countries, which is of significant concern.  From an Italian perspective, it is clear that current attempts to curb the pandemic are not in line with the Chinese experience. The Italians only instituted lockdown efforts when the active case load was close to 10 000, South Africa has made this leap at an active case load of 400. The next two to three weeks are critical in the evolution of the pandemic. Drastic and perhaps draconian attempts to arrest the disease must be made. Currently, taking the Chinese data at face value, all other countries are running below the Chinese active cases on a relative basis and can perhaps be contained. Again, the US growth rate is extremely high and of grave concern. In terms of the South African trajectory, we expect that if it follows the experience of other countries, there will be between 5 000 and 13 000 cases by the end of April 2020.

AAM updated this initial analysis as at the 25th of May; this is shown in Figure 5 below where we only plot active infections (UK has been excluded as there is no longer any data available on their recovery rate).

Figure 5: Comparative active infections of countries with elevated attack rates [5, AAM].

Figure 5 gives some interesting insight into some of the countries attempts to curb the extent of the viral spread; most of the European countries including Spain, Italy, France, Germany, Netherlands and Switzerland have managed to maintain their current infection rates below 100 000.  Several countries however are experiencing exponential growth in active cases (the US seems to be slowing but have 1.2 million active cases) significantly greater than any other country currently.  It would appear however that some of the emerging market BRICS economies are however struggling to contain their growth rates with Russia, Brazil, India and South Africa all currently exhibiting exponential growth.  Other countries of concern are Turkey, Peru and Chile.

Modelling the macroeconomic impact

The modern age has been defined by significant optimisation of capital usage. Balance sheets are run extremely tightly, working capital is often strictly efficient and inventories aligned with just-in-time principles. Global logistic chains have become the backbone of large corporations and small interruptions in these can have exacerbated downstream implications – the consequence of locking down portions of society to curb viral growth rates. The implication of this is likely to be a deep and extensive liquidity event. According to the National Bureau of Economic Research, the organisation responsible for the declaration of peak-and-trough business cycles, the current period of growth is the longest since records began in December 1854. These periods are shown along with the performance of the Dow Jones (in log scale) in Figure 6.

Figure 6: Recessionary periods as defined by the National Bureau of Economic Research for the United States since 1896[6, 7, 8, AAM].

The state of the economy is indicated as either -1 or 1 (cyan line) as a recession/expansion environment. The log return of the Dow Jones (blue line) shows the extent of the current sell-off. The theoretical return of being short of the Dow Jones in recessionary periods, showing the negative impact on the stock market, is additionally presented. It is clear that the number of recessions was significantly higher in the middle half of the last century. The fall-off is likely due to the introduction of monetary policy in 1933. Understanding the environment can assist an asset management decision framework significantly, especially from a tactical asset allocation perspective. The NBER, however, only declares these peak and trough inflection points between eight and 21 months after the event. The broadly accepted measure of two negative quarters in GDP is not considered. Rather, declines in economic activity captured by real GDP, real income, employment, industrial production and wholesale-retail sales are considered [6].

AAM has done significant research into the prediction of these periods using machine learning principles. A binary outcome decision tree model is trained using a typical in-sample cross-validation process. Reliable data was sourced monthly from 1939[7, 8]. To ensure that there is no overfitting, the data is split into two separate sets – this being before and after 1995. All cross-validation training is executed on the pre-1995 data set with out-of-sample performance carried out on the test data set, which is post-1995. It is critical that the model is able to predict the 2008 recession accurately to gauge confidence in the model’s ability. We consider several economic factors (known as features in machine learning). These include US employment, non-farm payrolls, industrial production, the three-month T Bill rate and 10-year constant maturity rate, corporate spreads and inflation. The process is run both with and without changes in the equity market. In this case (the S&P 500), the reason for excluding the equity market is that the NBER does not consider these moves in the pronouncement. Figure 6 shows the current expectation in terms of the expectation that at some future date the NBER will declare that the market activity had peaked and that the United States economy had entered a recessionary environment.

Figure 7: AAM internal prediction machine learning decision tree model, giving the current probability of recession [6, AAM].

As observed, the model is able to strongly predict a now cast of the state of the US economy and fits the periods of recession very well. This ultimately suggests that the machine learning protocol is able to accurately predict whether or not the NBER board will at some time in the future declare the current environment as being a point of inflection. To recap, the in-sample model prediction with (orange) and without (red) the equity market data is fitted to the NBER recession periods (blue). The out-of-sample prediction fits the two known recessions since 1995 well, both with (purple) and without (green) the equity market signal. Since equity markets are volatile, there is a tendency for a variance impact in the prediction in some cases. The current probability of a recession with the equity market included is roughly 20% and without the equity market only about 5%. The obvious caveat here is that the impact of COVID-19 has not yet borne through in the underlying economic times series and is likely to have a significant impact. AAM updated the economic model as at the end of May 2020; this is shown in Figure 8 below; we show only the inclusion of the S&P 500.

Figure 8: Updated AAM internal prediction machine learning decision tree model, giving the current probability of recession [6, AAM].

With the latest set of economic data it is clear that there has been a significant impact on the economic environment, with almost 100% probability that that US is now in an economic recession.  Currently the market is expecting an equally strong rebound in the economy towards the end of 2020; with significant fiscal stimulus allaying investors’ concerns on the performance of the global capital markets.

 One of the benefits of the methodology is that the mathematical gain of each feature can be extracted from the modelling exercise. Figure 9 below gives the importance of each factor in relation to its ability to accurately predict the economic environment.

Figure 9: Importance of various features in predicting macroeconomic now-casts [AAM].

The importance ranking of the features used points to the difference between the US 10-year rate and the three-month T Bills as the one that has the highest utility. This is a key indicator of the liquidity with negative values indicating tight market conditions. Figure 8 below shows this history.

Other features that are of value include the six-month and one-year changes in the equity S&P 500, changes in non-farm payrolls, changes in corporate spreads and changes in industrial production. All of these are likely to be impacted significantly in the future months. Early indications of manufacturing via the Philly Fed Manufacturing Index plunged 12.7% to its lowest level since July 2012, US jobless claims climbed to its highest level over the past two years, the S&P 500 is down over 25% and corporate credit spreads have moved out significantly. All of this points to an extremely high probability that the model will point with certainty that the US market will enter into a recession at the end of March 2020. To the eye it appears that most recessions are preceded by an inversion of the yield curve. This is indeed the case recently.

The potential offset to a large drawdown in the economic data will be the fiscal stimulus that most countries have or will introduce. Many countries have already cut interest rates and committed significant fiscal support for their economies. It is likely that markets will react positively to these injections of capital into the system in due course.

Figure 10: The 10-year US constant maturity bond rate less the three-month T Bill rate showing inversion of the yield curve since 1939 [6, 7, 8, AAM].

Conclusions

In terms of the COVID-19 virial infection currently sweeping the globe, it has been demonstrated that with concerted efforts and the lockdown of society, large growth rates can be curbed. There is no doubt that the courageous steps taken by Cyril Ramaphosa in locking down the country are warranted and it falls on the shoulders of all South Africans to observe their civic duty to enforce this period of isolation as assiduously as possible.  This appears to be supported by the latest trajectories with SA tracking significantly below other emerging markets currently.

However, it is critical that active cases peak in the next two weeks as demonstrated by China and South Korea. Even if this is achieved, the economic implications are likely to be severe. AAM will continue to monitor both the macroeconomic signals, as well as the trajectory of various countries fighting the pandemic. AAM will continue to use the tools developed to track both the epidemiological data, as well as the macroeconomic details. Please get in touch with the team if you require the ongoing analysis.

 

References

[1] Ross R and Hudson H, An application of the theory of probabilities to the study of a priori pathometry, Proceedings of the Royal Society, Volume 93, Issue 650, 1917

[2] Kermack W and McKendrick A, A contribution to the mathematical theory of epidemics, Proceedings of the Royal Society, Volume 115, Issue 772, 1927

[3] Peng L., Yang W., Zhang D., Zhuge C. and Hong L., Epidemic analysis of COVID-19 in China by dynamical modelling, Beijing Institute for Scientific and Engineering Computing, Feb 2020

[4] Hellewell j., Abbott S., Gimma A., Bosse N., Jarvis C., Russell T et al, ,Feasibility of controlling COVID-19 outbreaks by isolation of cases and contacts,, The Lanceet, Volume 8, Issue 4, Apr 2020

[5] Sourced from Johns Hopkins School of Public Health (https://data.humdata.org/organization/e5d3aa82-538e-4dae-94c9-010cc8ecbbc8)

[6] Sourced from the National Bureau of Economic Research, https://www.nber.org/cycles/cyclesmain.html

[7] Sourced from the Federal Reserve Bank of St Louis Economic Research, https://fred.stlouisfed.org/

[8] Bloomberg Data

[9] Sourced from Towards Data Science, https://towardsdatascience.com/how-quickly-does-an-influenza-epidemic-grow-7e95786115b3

 

Disclaimer

This communication/commentary (“this commentary”) has been prepared by Absa Alternative Asset Management (Pty) Ltd for information purposes only and must not be regarded as a prospectus for any security or financial product or transaction. It is not, nor is it intended to be, advice as defined and/or contemplated in Financial Advisory and Intermediary Services Act, 37 of 2002 (“FAIS”), or any other financial, investment, trading, tax, legal, accounting, retirement, actuarial or other professional advice or service whatsoever (“advice”).

While every effort has been made to ensure the accuracy of the above information, Absa Alternative Asset Management (Pty) Ltd does not accept any liability or responsibility for any loss, damage or expense incurred in relying on the above information or in the use thereof, nor makes any representation as to the accuracy or completeness of the above information.

Absa Alternative Asset Management (Pty) Ltd is an authorised Financial Services Provider with FSP number 22877.