What Drives G20 Equity Markets over time?  A First Pass at our review of Global Equity Markets through Data Science…

We routinely are inundated with news on minute-by-minute moves on stock markets, interest rates, commodities, the latest Central Bank announcement that has moved the market, what about inflation vs. deflation, etc.  Markets are a weighing machine as new information comes into play – but how does one measure the weight of each piece of information and does this weighing machine change over time? Markets are a very complex puzzle.

This is our inaugural post to review our research and models we have developed to date. We do not expect that future posts will be as detailed or as long.  

At Beowulf’s Treasury, we endeavor to provide fundamental research, insights, and education to help investors to slay their market dragon (emotional selling during unexpected drawdowns) and secure their treasure with proper risk management and quantitative tools to help an investor manage their emotions and behavioral biases.  

The Problem:

• The difficultly for most investors (both retail and institutional) is weighing the impact on market prices of new information or a change in expectations based on the new information is in constant flux which makes the puzzle very difficult to complete as the market is based on human behavior. Human emotions take over in highly stressful situations and are not always rational. Narratives or political agenda or bounded rationality all may skew an investor’s judgment as well.

• Due to the high level of indebtedness of the modern economy, financial markets have a significant impact on the business cycle. There appears to be global synchronization of the business cycle via the credit creation process given the interconnectivity of countries concerning the allocation of production and consumption activities. With a fiat currency system, the credit creation process has increased the level of indebtedness across households, businesses, and governments over time. At the end of each business cycle, debt service levels overwhelm the levels of cash flow as the wealth effect (“assets increasing in value enhance customer optimism increasing further consumption”) which leads to higher optimism leading to further credit expansion until the level of debt outstrips cash flow growth and debt serviceability. The only response to ensure the game continues to be played has been to reduce the debt service levels via interest rates and continue the flow of liquidity through all markets. This has benefited holders of real assets (stocks, real estate, etc.) over time. The short-term business cycle has historically lasted between 6-8 years.

• The current fiat currency/highly levered financial system is susceptible to periods of low volatility over extended periods, only to be interrupted by large unexpected volatility spikes and volatility tends to cluster which may result in large drawdowns (and market dragon if this results in emotional selling). Investors may follow the crowd and sell out of fear during these drawdowns as there is innate human behavior to avoid losses which are valued at nearly twice as great as the hope of earning a reward (known as loss aversion). Many investors may simply track price dynamics and high-level fundamental data without measuring the weights over time as they may not have the time or interest in tracking this data. Tracking the business/liquidity cycle and many of the other inputs in our BT model over time will help investors sidestep these costly drawdowns (known to us market dragons) by reducing equity market risk in decelerating conditions (so that you can secure your treasure), as there appears to be a repetitive pattern over time.  

A Potential Solution – Introducing the Beowulf’s Treasury (BT) Select G20 Machine Learning Model:

• Quantitative models may help identify these patterns, and based on our research these patterns tend to repeat over time. Through our process of building models, we understand cause-and-effect relationships to the best of our ability based on data rather than a narrative or emotion. This is critical in an ambiguous/uncertain environment and makes decision-making repeatable over time. The COVID-19 experience has been one of the most uncertain situations that many of us have faced during our lifetimes. It also highlighted the importance of quantitative modeling via projected infection rates as compared to hospitalization capacity, which led to informed policy decision-making and reduced uncertainty for the overall population. However, we do not blindly trust our models.

• We started this blog to provide our research, insights, and education on global markets based on free publicly available data. Within our research, we empirically test our hypothesis with quantitative models to determine how the market may be weighing new information and how factors may behave historically and in the future as well. This research may help us improve the probability of success when it comes to slaying the market dragon by attempting to sidestep large drawdowns.

Our Approach:

• We have a data-driven approach and built separate machine learning models for select G20 countries’ Equity Markets (as measured by national indices listed below in Table 4) and a non-correlated Safe Haven asset (the price of Gold in U.S. dollars). Table 1 below shows summary data on each of the countries captured in the study.  

• We were inspired to perform this study, after reading Ray Dalio’s recent book “Principles for Dealing with the Changing World Order – Why Nations Succeed and Fail” and previous book “Principles for Navigating Big Debt Crisis”. 

• We wanted to empirically test many of the theories and other hypotheses that other recent authors/pundits have proposed in the past to drive/lead equity markets. Through this testing and research, we believe our work may help investors position themselves for market turbulence and protect from large drawdowns through prudent risk management.

• For those readers that have not read Dailo’s most recent book, the central theme is that there is a certain shelf life for the world’s main reserve currency based on the past 500 years of recorded financial history and follows a similar arc over time, eventually leading to over-indebtedness and may result in a conflict between the declining existing power (reserve country) and a country that is an emerging power and/or an internal conflict (civil war). These conflicts are generally based on disagreements about how to allocate and divide the world’s resources (inclusive of wealth inequality). The perils of greed (accumulation of wealth over lifetime consumption, without redistribution) and overconfidence are common themes within the epic poem Beowulf which inspired us – more on this in the About Us section of the website.    

• The reserve currency countries have historically lasted between 75-100 years and these conflicts tend to disrupt the existing world order. Since WWII we have largely been able to keep large-scale world conflict to a minimum as a result of mutually assured destruction. Trade, innovation, and economic prosperity have improved the lives of many globally. As a consequence, debt and asset prices have grown over time as there is a confidence that the future may be better than what we see today through enhanced productivity and innovation. Historically, these worldwide conflicts tend to redistribute the world’s resources including normal households that may be saving for future consumption or investment.   

• This pattern has played out via the Dutch Guilder, British Pound, and now potentially the U.S. dollar and potential conflict with an emerging power in China. A longer-term objective of this blog will be to monitor the central theme of the book, is the time of the U.S. dollar as a world’s reserve currency coming to an end? What currency may replace it?

• In our first pass BT model, to help us answer this question over time, we integrated monthly uncertainty measures such as the Economic Policy Uncertainty (EPU) and Geopolitical Risk Indices (GPR) provided by www.policyuncertainty.com and Partisan Conflict Index published by the Philadelphia Federal Reserve which tracks the degree of political disagreement among U.S. politicians at the federal level.  These indices are based on news-based measures of policy uncertainty, adverse geopolitical events, and internal political division. We have observed the increasing risk of a U.S. internal conflict through January 6^th insurrection and protests during the last few years and these types of events would be reflected in these measures.  

• Market participants and central bank officials view economic policy uncertainty and geopolitical risks (inclusive of internal political division) as key determinants of investment decisions and stock market dynamics. So the rate of change over time will give us a data point in our model to track the potential for conflict.  

• Before jumping into the models, you may ask why cover the G20? The G20 was founded in 1999 in response to several world economic crises. These countries have collaborated further to construct the financial infrastructure of the global financial system today (i.e. Basel III bank regulations) which underpin the creation of credit via capital and liquidity rules. These countries account for a significant amount of the global equity market cap, private debt, and total reserves held by central banks. These countries have also attempted to coordinate their commitments on Climate change. By following these countries we get a better sense of the global liquidity cycle influencing the business cycle, and stock prices globally.     

• The BT Machine Learning models attempt to capture underlying fundamentals of the economy, market sentiment/investor positioning, short-term debt cycle also known as the economic cycle which is determined by liquidity and credit from the banking system and actions of the government (fiscal and monetary policies), as well as the structural themes which Dailo’s book covered in detail.

The Model Details:

• We compared 12-month log changes against several features across 7 categories across Investor Sentiment, the Banking/Credit Environment, Growth/Inflation Cycle, Macroeconomic/Geopolitical (Structural), Household Finance/Consumption, Market Sentiment, and Other/High-Frequency data which are largely consistent across each of the countries reviewed. For each country model, we have gathered 50 to 60 features/indicators across the above 7 categories (see Table 3 below) to train and fit our models. The information has been either disclosed by the respective country’s Central Bank, National Statistical Organization, Securities Exchanges, or global organizations such as the Organization for Economic Co-operation and Development (OECD) or other academic organizations. To the extent possible, we attempted to automate the data collection process so that we may quickly update our findings at a later date. Table 2 below illustrates the model infrastructure and training and testing splits that were followed for each of the country models.

• The features used in the models are mainly leading and coincident indicators. These feature sets are used as training data to fit each model. Training a model involves using an algorithm to determine model parameters (e.g., weights) to map inputs (independent variables) to a target (dependent variable – rolling log 12-month change in Share Prices).

• A horserace of 12 models was trained with the objective of minimizing an error function which was the mean squared error (MSE). On average, 18 years of data (on a monthly basis) were used to train the models (using Python Jupyter Notebook).

Table 1 – Selected Features – Select G20 Countries

The models were tested out-of-sample for an average of 5 years. We averaged the results across the top 5 performing models which are both linear and non-linear regression models (decision tree and random forest models). 80% of the data was used to train the models and the remaining 20% of the data is used as out-of-sample testing.  Test data was not used to build the models. The duration of the training set of data is typically limited by the availability of data. For example, Margin Debt reporting only began in Canada, Korea, and Australia in the early 2000s. An earlier version of these models started in 1991 and results were similar to the output below.

Table 2 – BT Select G20 Model Infrastructure and Training/Testing Split

So what did we find? Are these models predictive?

The short answer, the first pass results we believe are promising…however this is the first pass. Table 4 below summarizes the summary model output.    

Table 3 – BT Select G20 Model Feature Set and Features with High Model Value (in Green)

Table 4 – Summary of BT Select G20 Model Machine Learning Models

Table 5 – Summary Economic Cycle (Share Prices, Liquidity, Credit Impulse, and Select G20 OECD CLI)

Table 6 – Correlation of Equity Markets based on rolling 12-mth Returns (Daily Prices) – 2013 to 2021

BT Model Observations:

The models have held up well in the out-of-sample testing period of the previous 5 years (2017-2021) with an average of R-squared of 0.76 (goodness of fit of actual returns versus predictive modeled returns based on test data) and Root Mean Squared Error of 0.10 in the out-of-sample period.  

• Models with longer data sets across multiple business cycles have performed relatively well across time (the U.S. model – S&P 500). Given the depth and maturity of the U.S. capital markets, high-quality information on a more frequent basis on a monthly or quarterly basis (more on this model later)…     

• Growth/Inflation Cycle: OECD Composite Leading Indicator (CLI) is a key factor in all country models and appears to pick up turning points in stock markets well in advance of major drawdowns, as well as, economic turning points. The composite leading indicator (CLI) is designed to provide early signals of turning points in business cycles showing fluctuation of the economic activity around its long term potential level and appears to be a good proxy for GDP given the lag in reporting of quarterly GDP (typically we have to wait 2 months after quarter-end for these numbers). High-frequency monthly production data such as Plastic production and Electricity consumption appear significant in China and Mexico (countries that rely more on production for economic growth on a relative basis).    

• Banking/Credit – The rate of change in Global Money Supply/Liquidity and Credit Impulse appears to have an impact on the economy, as well as, returns in equity markets over time. This is significant to note as currently as the rate of change in global liquidity is decelerating and appears will continue to do so through balance sheet management and interest rate normalization, which is a major risk to equity markets in the near term. Table 5 shows there appears to be a certain amount of synchronization of the global liquidity cycle. European equities appear significantly sensitive to the overall global liquidity cycle based on the BT models.      

• Investor Behavior – Household Balance Sheet Allocation to Equity Markets: The rate of change in Household allocation to equity markets relative to fixed income implied by Financial Account in the Flow of Funds and Margin Debt features tends to perform well across all markets in which the information is available.  Though it is hard to directly observe in the data, there appears to be some home bias (domestic investors holding a high proportion of their portfolio in domestic securities) given the strong correlation between household balance sheet equity allocations and home markets.

• Investor Behavior – Trading Margin Debt Balances: With adopted of quantitative easing, investors have increased their risk appetite as evidenced by higher allocation to equity markets and higher margin debt. As markets have grown over time, margin balances (where disclosed) have expanded. Changes in margin balances are correlated with changes in market values

Table 7 – Z-Scores: Share Price, Household Equity Allocation and Margin Debt – Rolling 12-mth change

12-mth change in Share prices, Household Allocation, and Margin Debt have generally stayed within 2 standard deviations over time supported by the credit/liquidity cycle. Current conditions across most markets show a deceleration 12-month share price change, Household Equity Allocation, and Margin Debt (see Table 7).  

• Commodity Driven Economies: Countries that have significant activity related to mining resources have a significant correlation to changes in industrial metals such as Copper (Canada, Russia, Australia, Brazil). The Copper/Gold Ratio also tends to track the OECD CLI well over time and this could be a significant cyclical indicator that you can monitor daily (more on this later)….  

• Growth/Inflation Cycle: Corporate Profits and Liquidity: Earnings-per-share at the index level and Corporate Profits within GDP are significantly correlated and importance to GDP over time has become more significant as policy frameworks since the 1980s have generally favored capital relative labor across most G20 countries. Also, the rate of change in the Money Supply/liquidity and Corporate Profits/EPS appears to be strongly correlated. We note that companies have also grown reliant on the ongoing flow of ample liquidity since the Great Financial Crisis (GFC): a growing cohort is borrowing to deploy buybacks and support earnings-per-share through financial engineering. With central banks poised to normalize rates and reduce liquidity, the second cohort of companies that are cash flow negative and relying on ongoing new investments to finance their business is at risk. With less money flowing into the system, that funding will be harder to come by. We may be able to observe the initial smoke related to such a situation via widening of lower quality credit spread (a feature in the models). This may be a topic for further research in due course.

Macroeconomic/Geopolitical – Outperformance in Economic Growth has not Translated to Share Price Outperformance:

• China’s entry into the World Trade Organization in 2001, has become the world’s largest exporter via investment in production capacity, utilization of natural/labor resources becoming a net creditor, dethroning the U.S. as the world’s largest trade partner. The rest of the world has benefitted from this increase in the global labor force (benefitting from a demographic dividend) which has been a significant deflationary impulse across all economies over the last 15-20 years. The People’s Bank of China (PBOC) has become the largest central bank reserve assets holder with almost 40% of total central bank reserve assets globally. However, despite the strong economic performance, this has not led to outperformance in equity markets on a relative basis.
• Domestic allocation to equity markets appears low on a relative basis as roughly 10% are allocated to financial markets and 70% of Household Assets are allocated to Real Estate. By comparison, U.S. households hold 40% of assets in equities and 30% allocated to Real Estate.  Also, foreign portfolio investment positions continue to be low in China relative to the size of the economy and household assets, as well as, more mature capital markets such as the U.S., EU, and the UK with a higher foreign investor base (foreign portfolio investment is >50% of the money supply). China has gradually opened its capital markets over time to foreign capital has improved and globally held ETFs have expanded their allocation to China over time. However, China’s vast yet protected home market has allowed some of its domestic firms to acquire a scale that provides them with significant advantages when they compete in other global markets. China has been particularly protective of strategic industries such as technology (including last year’s crackdown on domestic technology firms and their founders), given the importance to future economic growth and maintaining existing governance structures. Both the United States and China are taking measures to reduce the leakage of sensitive technologies, and we expect scrutiny of foreign investment to continue given the strategic competition between the U.S. and China. This is a theme we will continue to monitor.  

• Prioritization of real estate relative to financial assets by Chinese citizens, coupled with strong credit creation since 1999, could be one of the reasons that Chinese Real Estate has become one of the largest asset classes in the world. However, since 1999 financial markets in China have been slowly liberalized, encouraging capital inflows from foreign investors, while keeping domestic social stability in check and encouraging domestic financial literacy. 

• However, for the yuan or any other currency including crypto to displace the U.S. dollar as the world’s leading reserve currency, the yuan must take on a larger role for capital raising and investment and trade/payment system. China has been taking steps to reduce its reliance on the U.S. dollar payment system by introducing a Central Bank Digital Currency (CBDC) which may eliminate its dependence on the U.S. dollar payment system. Also as China attempts to move from a production-led export economy to focus on the development of domestic market/consumption, we will be watching the performance of the model over time. Table 8 shows the evolution of Global Central Bank Foreign Exchange Reserves.  At this point, though the U.S. has lost ground since 1999, central banks hold roughly 60% of foreign exchange reserves in U.S. dollars.

Table 8 – Global Central Bank FX Reserve – the U.S. Remains on Top for Now

• Despite increasing fiscal deficits/debt, trade deficits, and lower GDP growth,  the U.S. dollar continues to be the world’s reserve currency in which goods are quoted and traded, and with which payments are settled (particularly in global commodity markets). This has provided the U.S. the exorbitant privilege of borrowing trillions of dollars from the global capital markets in U.S. dollars without facing a significant balance of payments crisis/depreciation of the dollar, at minimal interest rates and with virtually zero default risk. Also, other net creditors may have excess dollars over above their fixed income holding requirements at the current rate which drives these excess dollars into U.S. financial assets including equity markets.

• China’s foreign currency assets composition is not disclosed but has been suggested that it approximates the overall global composition of ~60% of USD, ~20% Euro, GBP ~5%, and JPY ~5% given the relative size of its overall holding of central bank reserve assets at 40% of the total.  Both Russia and China have been trying to reduce their reliance on U.S. dollar trade (including a higher allocation to gold as part of central bank reserves) as they currently rely on trade in U.S. dollars and sanctions by the U.S. government may be placed which may disrupt the flow of trade. China is poised to continue to move from producing/export-led economy to consuming/service economy, in which stronger currency and continuing broadening of access to capital markets becomes important, coupled with enhanced domestic financial literacy and a higher allocation of household savings to financial assets.

• Historically when the devaluations (via increases of money relative to real assets such as stocks, real estate, gold) have become extreme that the financial system breaks down. In response, governments are forced to back the money with a physical form of hard currency (typically a limited resource such as gold) to rebuild the trust and faith in the currency within the financial system so that the value of the underlying money is not volatile and maintains its store hold of wealth.  Since 1999 gold reserves have represented ~10% of central bank reserve assets of the select G20 countries and have increased relative to the broad money supply from 0.2% to 1.0% in 2021 (see Table 9).       

Table 9 – Gold as % of Broad Money Supply of Select G20 Countries

• Demographic trends in China are expected to put pressure on economic growth over the next decade based on UN projections (without significant immigration), the population will shrink significantly from current levels. We have seen similar trends play out in Japan before during the late 1980the /the early 1990s as well.
• We plan to dig into this demographic theme in a future report given that over the next 10-15 years there is an expectation that Brazil, India, Mexico, China, and Indonesia are expected to make up a larger proportion of the global middle class. As a result, these economies may focus more on the development of consumer-led consumption and more reliant on the ‘Wealth Effect’ to grow the economy which may substantially change how production and consumption are allocated globally, and potentially further development of the domestic capital markets. As an example, recent ‘Common Prosperity’ measures in China focus on reducing wealth inequality over time to spur domestic demand, increase self-reliance, reduce the reliance on real estate/credit growth, and create more balanced economic growth going forward. These measures may enable greater wealth to be invested by the middle class in equity markets and spurn further innovation over time. This is expected to reduce internal conflicts over resource allocation as well reducing economic policy uncertainty.

• High levels of Foreign Portfolio Investment (openness to foreign capital and trade, which has historically driven innovation and growth), coupled with High Domestic Household Allocation to the domestic Equity Market appear to be the winning combination for the U.S. for now. This may be supported by strong earnings growth and lower volatility in earnings on a relative basis. This may also be the fact that the U.S. dollar is the world’s reserve currency which has run twin deficits (trade and fiscal) and net creditors such as China, Germany, and Japan must recycle US dollars back to buy financial instruments (bonds and equities).   

• Household Finance/Consumption: In highly financialized economies, the ‘Wealth Effect’ in which higher asset prices (Real Estate Wealth and Stock Prices) cause households to be more confident in the economy and spend more appears to be more evident in China, the U.S ., EU, UK, Canada, South Korea, and Australia based modeled outcomes. Real Estate prices have tended to trend up along with Private sector Credit-to-GDP ratios. Banking regulations support the ‘Wealth Effect’ through housing prices given that residential mortgages which are secured by collateral generally receive a relatively low-risk weight given that probability of default and loss given default tend to be the lowest given high-quality collateral, which in turn may provide a bank a higher return on equity relative to relatively risker loan such as a business loan. Also, capital regulations treat sovereign bonds on a preferential basis at a 0% risk weight. Liquidity regulations require large cushions of High-Quality Liquid Assets (HQLA) to be held by banks, which are mainly sovereign exposures. Most banks tend to have the bulk of their sovereign portfolios in debt of their home country given the preferential regulatory treatment and banks are significant investors in sovereign debt. We witnessed this type of behavior during 2020 as increased client deposits via payments sent from the government are backed by sovereign bonds held in the bank’s liquidity portfolio. For the ‘wealth effect’ to continue, new liquidity/credit must enter the economy through credit creation or payments from the government to mitigate disinflation which is why the credit impulse and liquidity growth is an important metric to track going forward.
• It should be pointed out that since the GFC, the countries at the center of the banking crisis driven by the collapse of a real estate bubble (U.S., EU, and the  U.K.) have been able to slow the credit growth to the private sector and improve the overall balance sheet partially through higher fiscal spending via debt monetization. Credit to the private sector in China, Canada, Korea has significantly expanded over this time.  Going forward, the composition and evolution over time of private versus public debt are worth watching, as historically the public sector which can create money tends to assume private sector debt during a crisis or recession to avoid a collapse of the financial system.   

BT Model Review – What surprised us?

• Leading indicators for GDP such as the OECD Composite Leading Indicator performed extremely well across all jurisdictions. Careful monitoring and positioning (exiting the equity market when 12-mth change falls below 0%) may help protect investor portfolios from significant downturns.

• Earnings per Share/Corporate Profits, Yield curve/interest rates were not as significant as expected. The short-end of the curve (changes in the 3mth rate) appear to have the most impact on model output.  

• Risk Appetite/Investor Behavior is a very significant factor in the results as measured through Household Balance Sheet Allocation to Equities (Flow of Funds position Accounts) and Margin Debt were significant across developed nations where interest rates/economic growth is low.

• Despite a relatively long time series for model training, the South Korea BT model did not perform as well out-of-sample period (low R-squared, high RMSE) as other countries. The model performed well in the out-of-sample period between August 2017 to June 2021. From June 2021 to October 2021, actual returns were higher by at least 10% on a rolling 12-mth basis than predicted by the BT Model.

• Geopolitical Risk Index did not have a significant impact on BT Gold model in training and out-of-sample period, however, the GPR has been relatively range-bound throughout 1999 to 2021 (ex-Sept 11/2001 Terror Attacks).  Generally, during a crisis, there is a flight to safety (U.S. Treasury bonds, gold, and the Japanese Yen).

What are some of the limitations of the BT Machine Learning model?

• We may not have chosen the right independent variables/predictor variables of the overall stock markets of the select G20 economies. Significant pivots in the economic sectors of the economy and stock market indices and/or energy transition from fossil fuels to green sources over a short period may not be fully incorporated into the model without further expansion of the data set.
• Statistical relationships of certain predictor variables may change over time given that markets are highly influenced by human behavior and production and consumption allocations may change over time.
• The fact that an independent variable turns out to be significant says nothing about causality. Certain factors may be significantly intertwined and difficult to separate from each other. Also by their very nature, financial or economic outcomes/data may not be normally distributed and may exhibit a non-normal distribution such as fat-tails (high kurtosis, low probability of high impact events).  This is an especially relevant point concerning observational studies.

• We have tried to the best of our ability to gather comparable data from each jurisdiction. However, data availability is not consistent across the countries (e.g. Climate Policy Uncertainty Index and Partisan Conflict Index only exist for the U.S. at this point) within this study which limits the amount of data that can be used to train the machine learning models. We noted above those models with a longer time series of data generally performed better in out-of-sample testing.

A Deeper Dive into the U.S. Model (S&P 500): 

• The rate of 12-month change in U.S. markets appears to be strongly correlated with the rate of change with an R-squared of 0.90 (goodness-of-fit) in the out-of-sample period 2013-2022:
  • OECD Composite Leading Indicator for the U.S. (a proxy for GDP)
  • U.S. Short-term interest rate (3-mth) – a proxy for Fed Funds – 3 rate hikes in 2022;
  • U.S. dollar liquidity (Money  Supply);
  • U.S. Margin Debt/Household Allocation of Net Wealth to Equities (Risk Appetite);
  • Domestic Consumer Confidence;
  • Domestic Credit Impulse;
  • High Yield Corporate Credit Spreads;
  • ISM/ISM New Orders;
  • Earnings-per-share;
  • Personal Income/Retail Sales; and
  • Housing Starts/Real Estate Prices.

• These factors are consistent with an economy that is driven by consumption (retail sales, housing, vehicle sales, etc.) which has the world’s leading reserve currency and is as a result of twin deficits (trade and fiscal) within a mature financialized economy with a population of high financial literacy on a relative basis.  

• Based on 2022 consensus estimates for Corporate Earnings/EPS of the S&P500 (expected up 9% YoY) and the economy from Survey of Professional Forecasts Philadelphia Federal Reserve, based on the BT model it is expected the S&P 500 will end the year up 5% from 2021. This is consistent with the decelerating trend we are experiencing in liquidity, credit impulse, equities allocation, and margin debt.

Table 10 – Out-of-Sample Testing U.S Model (S&P 500) – Model Prediction vs Actual

Table 11 – Other Equity Market Out-of-Sample Results – Model Prediction vs Actual

A Deeper Dive on the Gold Model:

• The rate of 12-month change in the gold price (in $US dollars) appears to be strongly correlated with the rate of change with an R-squared of 0.88 in the out-of-sample period 2018-2021 along with the following features:

• Central Bank Gold to Total Reserve Asset Ratio;
• Select G20 Total Reserve Assets;
• The U.S. Household Balance Sheet Allocation to Gold relative to Stocks and Bonds;
• U.S. Real 10Y/30Y interest rates;
• Price of Oil and U.S.CPI;
• Select G20 BT Credit Impulse/Commercial & Industrial Lending in the U.S.;
• Safe Haven Asset (Japanese Yen/weakness in USD(DXY)); and
• U.S. Economic Policy Uncertainty measured by the EPU Index.

Table 13 – BT Gold Model – Out-of-Sample

Introducing the Daily BT Risk Indicator: 

• The OECD CLI is a significant indicator based on our research to track going forward, however, there is a lag of about 45 days to report results for a given month. This makes our approach susceptible to quick market drawdowns, such as the one we experienced in February/March 2020. However, looking back during that time the OECD CLI was already in negative territory and would have indicated an extremely cautious risk position in our investment portfolio.
• To manage and monitor our risk on a more frequent basis, we have attempted to create a customized index that may be tracked daily that aligns well every month to the select G20 OECD CLI (see Table 14 below). This custom index which we have called the BT Market Risk Indicator which includes the following measures:
  • Copper-to-Gold Ratio
  • High Beta-to-Low Beta Currency Pairing 
  • U.S. Discretionary Consumer Spending-to-Consumer Staples Ratio
  • U.S. High Yield Corporate Credit to Intermediate Gov’t Bond Ratio
  • U.S. TIPS Bonds to Intermediate Gov’t Bond  Ratio

Table 14– BT Market Risk Indicator

Potential Further Research Topics:

• The first pass at the model introduces many further research questions that we hope to provide updates over time: 
  • Inclusion of Forecast for 2022 and beyond based on consensus estimates (where available) to determine expected returns for each index and potential portfolio allocation for informational purposes.  

• Determine if we may be able to further calibrate models and remain predictive out-of-sample. Review Feature sets for completeness. Based on reader interest/feedback, we may update our analysis from time to time.  

• Deeper dive on each equity market composition by sector and understanding how this may tie into the factor analysis. Also extending the models to determine if there are any insights on a cross-sectional basis (i.e. the impact of U.S. consumption on large exporting nations such as China).

• We have scratched the surface on the global liquidity cycle in this post, a deeper dive on underlying sensitivity to liquidity conditions by index may be warranted. 

• Additional models developed for other equity indices (NASDAQ, Russell 2000), other G20 countries (South Africa and Indonesia), Commodities models (i.e. Copper price and further discussion on Copper-to-Gold ratio), and potentially other asset classes including interest rates and cryptocurrency. Review the potential for inclusion of the Taiwan Stock Exchange and other related data given the strategic importance of semiconductors related to the U.S./China relationship going forward.     

• Daily tracking of BT Risk Indicator for real-time Global OECD CLI

If you have gotten this far, we appreciate your focus and thank you for reading our inaugural post. We do not expect that future posts will be as detailed or as long. Appreciate any questions or feedback you have or other research topics you may suggest.

Please contact us at beowulftreasury@gmail.com.

BT