Simple Macro Regime Framework For Portfolio Construction

Share:

Classical portfolio construction methods like mean-variance optimization are infamously very sensitive to input assumptions (ie asset class’ expected returns and covariance matrices) which vary significantly through time. Drawing heavily form Ray Dalio’s All Weather Portfolio and frameworks eloquently popularized by 42Macro among others, we aim to reconstruct a “poor man’s” version of the approach. We seek a simple and repeatable methodology to inform portfolio allocation and rebalancing timing with a view to outperform passive allocations in terms of risk and returns.

In this note, we define simple macro regimes based on CPI and CLI’s rate of change and explore a “proof of concept”, unlevered, long-only, 3-asset portfolio constituting of S&P 500, long dated Treasury bonds and gold positions only.

We will subsequently look to:

  • Broaden the investable universe to over 15 asset classes (at the expense of back-testing history) across equity style factors, geography, duration, credit and commodities
  • Explore additional input variables in the hope of better anticipating regime changes changes ahead of CPI and CLI publications

Macro regimes definition

Macro regimes are defined by the rate of change in inflation and real growth as opposed to their levels, that is to say the second order derivative of real GDP and price levels. For example, a YoY inflation print going from 5% to 4% from one month to the next is defined as “deflationary” even though prices are obviously still rising. Similarly a real growth rate going from -2% to -1% (or equivalent measure) will be considered a positive impulse to real growth even though the economy is still contracting.

Initially, we ignore the magnitude of those change or their divergence from surveys, forecasts or market expectations and focus only on the direction of their rate of change. We further leverage 42Macro’s elegant nomenclature and define the following 4 macro regimes:

  • Growth (or real growth, “G”): real growth accelerates while inflation decelerates
  • Reflation (“R”): both real growth and inflation accelerate
  • Inflation (or stagflation, “I”): growth decelerates, or declines at a faster pace, while inflation accelerates
  • Deflation (“D”): both growth and inflation decelerate

The framework is initially implemented using US only monthly data from 1972 to 2022.

  • Growth regime is based on OECD amplitude adjusted US CLI month on month change
  • Inflation regime is based US CPI YoY rate’s change month on month

The above definition nicely splits the past 50 years in 4 roughly equally frequent buckets with only a slightly larger likeliness for inflationary regimes I and R. The relative frequency of each regime also appears fairly homogenous in time, i.e. the US economy spent the same relative amount of time in each regime through the past 50 years as it did in the past 25 years and 15 years (see pie charts below).

While notable periods of expansion and contraction are visible, the regimes are pretty “noisy” and flip from one to another a lot. Smoothing CPI and CLI data using moving averages and longer lookback periods (e.g. 3 month change rather than month on month) help mitigate that noise but not by much. For the sake of simplicity and repeatability (limit the number of data massaging steps), we stick to month on month change definitions.

US CLI, YoY CPI and macro regimes since 1972. Source: Bloomberg OEUSKLAC Index and CPI YOY Index
Stable frequency distribution of each regime in the past 50, 25 and 15 years

Single asset performance by regime (stocks, bonds & gold only)

Limiting our scope to stocks, long bonds and gold for now, the below table shows asset class monthly returns since 1972 by macro regime. Far from any surprising or breathtaking conclusions, it is still nice to validate intuitive and established asset class features:

  • Stocks perform better than average on both a risk and return basis during expansionary regimes (G & R) . In particular in G regime, average monthly return is 2.3% (27% annualized) compared to 0.9% (11% annualized) overall and monthly stock returns are positive 75% of the time compared to 63% overall
  • Bonds outperform during Deflation (D) regimes and underperform during Reflation (R, D’s “opposite”) as positive inflation and growth impulses are generally associated with parallel shift up and steepening of the yield curve. G and I regimes show no statistically significant difference in performance as growth and inflation’s impact on yield offset each other.
  • Gold, while not statistically significantly different from its overall average performance, does show signs of outperformance in inflationary regimes (R & I)
Asset class return by macro regime and statistical significance

3-asset portfolios

We draw the efficient frontiers of all¹ possible long-only, 3-asset portfolio for each GRID regime as well as for the entire combined 50-year history. The below graph also highlights the location of the traditional 60/40 portfolio and the “optimal” portfolio as defined by the max Sharpe ratio².

When optimized across the entire history (grey line below), the efficient frontier is relatively flat. This is no doubt partially attributable to our limitation to only 3 assets, however, given the global efficient frontier (grey line) is nearly completely dominated by the regime specific ones (colored lines), this suggests diversification benefits occur across macro regimes rather than within them

Consistent with that observation, the regime-specific optimal portfolios (colored circle) are fairly concentrated:

  • Expansionary regimes’ (G & R) optimal portfolio are 80% stocks only diversified to optimal risk / reward by 20% bonds in G regime (real growth) and 20% gold in R regime (reflation)
  • Optimal portfolios during contractions are dominated by bonds with a 100% allocation during D regime (deflation) and only a 20% diversification to gold in I regime (inflation / stagflation)
  • The results echo the, again very intuitive and well known, single asset performance described previously where stocks and bonds dominate in expansions and contractions respectively and gold helps absorb relative pressure from inflationary regimes (R & I). Despite the obviousness of the result, its consistency with actual data is remarkable

Finally, it is worth noting the traditional 60/40 portfolios only appear on the efficient frontier in the G regime and the overall sample. Within R, I and D regimes, the 60/40 portfolio offers sub-par risk / reward

Efficient frontiers of 3-asset portfolios in each macro regimes. (stock%, bond%, gold%)

Each regime’s optimal portfolio displays different levels of risk (as measured by monthly standard deviation). In order to carry the same amount of market risk across regimes, we normalize final portfolio allocations to a 2.5% standard deviation (the minimum of the 4, consistent with our unlevered constraint). Optimal 3-asset portfolio compositions are summarized below along with their risk normalization.

Regime-specific optimal portfolio composition and risk normalization

It is further worth noting that the global optimal portfolio (30% stocks, 50% bonds, 20% gold) is pretty close to the simple average of the regime-specific allocations. This is due to the occurrence of each regime being almost uniform as observed previously. In other words, the global optimal portfolio would be chosen absent any indication of the prevailing macro regime (i.e. c. 25% chance of each).

Finally, the global optimal is directionally in line with vanilla versions of the All Weather portfolio and its signature over-allocation to bonds and inflation protection (gold here) compared to commonly expected stock-heavy tilts.

Regime transitions

All asset and portfolio monthly returns have been presented assuming perfect knowledge of the prevailing macro regime at the time. Regime definition relies on lagging data and both CPI and CLI data for a given month are published in the first 2 weeks of the following month. Our monthly data granularity is too coarse to capture intra-month variations before and after actual publication. Strictly speaking, at the start of each calendar month, we only have complete information about the macro regime 2 months prior.

We will explore macro regime forecasting more holistically at a later date, for now, we limit ourselves to a naïve observation of regime transition probabilities from one month to another. Assuming we have previous period data available, and therefore know the previous month’s regime, we will hold a blend of the regime-optimal portfolios per the probability of transitioning to that regime from the previous period… That was a mouthful, see “1 step transition” in the table below:

  • If we are in March and February regime is known to be G. based on historical transitions, we expect a 45% chance of March remaining in G, 45% chance of transitioning to R, 3% chance of transitioning to I and 7% chance of transitioning to D
  • Our March allocation is a pro-rata portion of each regime’s optimal portfolio per those historical expectations. For example, regime-optimal stock allocation for G, R, I, D regimes are respectively 70%, 65%, 0%, 0%. As a result, blended by 1 month transition probabilities, the stock allocation becomes 45%.70% + 45%.65% + 3%.0% + 7%.0% = 61%
  • We repeat the same blending for 2 months transition
Regime transitions and blended portfolios

Back-test and benchmark

The performance of the defined optimal and blended portfolios are benchmarked using monthly asset return data since 1972.

The “Ideal” strategy models a monthly rebalancing to the appropriate regime-optimal portfolio assuming complete knowledge of the current month’s macro regime. While unachievable in practice given the perfect foresight assumption, this provides a view on the approach’s upside potential should macro regime forecasts be improved.

The “Blended lag 1 & 2” strategy models only assumes knowledge of the previous 1 or 2 month’s regime and rebalances to the appropriate blended portfolio defined above. This constitutes a practical and repeatable base case range for the framework and the proof of concept we set out to establish using only 3 assets.

The performance is benchmarked against the global optimal portfolio (implicitly ignoring any regime information), a passive 100% stock allocation and a monthly rebalanced 60/40 portfolio.

All calculations ignore taxes, slippage or transaction costs.

Return on $1 invested

“Blended — lag 1 & 2” portfolios, which again are implementable in practice, outperform 100% stocks and 60/40 portfolios in the long run both in terms of risk and return (higher CAGR, larger Sharpe, lower drawdown). While some decades favor stocks in terms of raw returns (1983–92 and recently 2013–22 offering 4.3x and 3.6x MOIC respectively), the risk tradeoffs remain in favor of the blended portfolios (c. 50% higher Sharpe)

Portfolio statistics

Conclusion and next steps

This “toy example” 3-asset portfolio is convincing enough that this macro regime framework can offer significant risk and drawdown reduction while providing returns comparable or better than stocks in the long run. Natural extensions to further improve performance include:

  • Expanding the investable universe beyond the 3 assets considered here. In particular, we expect equities sector and geography diversification as well as the inclusion of credit assets can enable larger sustained stock allocations through different regimes. Furthermore the inclusion of TIPS, commodities and real-estate should damper gold’s volatility while sharing its inflation protection features during
  • Improving regime transitions and blended portfolio definitions. The approach presented here is pretty naïve and additional analysis might allow better forecasts than those provided by the simple historical transition probabilities used here. Any incremental precision should bring the strategy’s performance closer to the “Ideal” portfolio

1: To be exact, we use a coarse 10% mesh to define “all” portfolios
2: This a loose use of Sharpe ratio which we use to designate the ratio of average monthly returns to monthly standard deviation with no annualization or risk free rate assumption