HomeMacro TrendsMacro demand-based rates strategies

Macro demand-based rates strategies


Jupyter Notebook

The pace of aggregate demand in the macroeconomy exerts pressure on interest rates. In credible inflation targeting regimes, excess demand should be negatively related to duration returns and positively to curve-flattening returns. Indeed, point-in-time market information states of various macro demand-related indicators all have helped predict returns of directional and curve positions in interest rate swaps across developed and emerging markets. The predictive power of an equally weighted composite demand score has been highly significant at a monthly or quarterly frequency and the economic value of related strategies has been sizeable.

The below post is based on proprietary research of Macrosynergy.

A Jupyter notebook for audit and replication of the research results can be downloaded here. The notebook operation requires access to J.P. Morgan DataQuery to download data from JPMaQS, a premium service of quantamental indicators. J.P. Morgan offers free trials for institutional clients. Also, there is an academic research support program that sponsors data sets for relevant projects.

This post ties in with this site’s summary of macro trends and systematic trading strategies.

The basic theory behind aggregate demand and interest rates

The level of interest rates in an economy affects aggregate demand in the macroeconomy, particularly private consumption, fixed investment, and inventory changes. All other things equal lower real (inflation expectation-adjusted) interest rates make savings less attractive, borrowing less expensive, and, thereby, lessen the outright and opportunity costs of purchasing goods and services. In particular, interest rates raise the affordability of durable goods purchases by households and lower the barrier of profitability for investment projects. For a macroeconomic equilibrium this has been formalized by the IS curve, a negative relation between the output gap and the real interest rate. For a somewhat broader context see our previous post on simple macroeconomics for trading.

Conversely, the pace of aggregate demand relative to a sustainable or desirable pace is indicative of macroeconomic pressure on interest rates as an equilibrium price. Strong or excessive aggregate demand exerts upward pressure, while weak or sub-standard demand puts downward pressure on real interests. Considering the rational inattention of markets, efficient tracking of demand trends is therefore a valid guide for duration exposure, for example in the form of fixed receiver positions in interest rate swaps.

The pace of aggregate demand is also an important factor in the direction of monetary policy. All other things equal strong aggregate demand calls for tighter monetary policy and weak aggregate demand supports the case for monetary easing. In a credible inflation-targeting regime monetary tightening leads to a flatting yield curve and monetary easing leads to a steepening curve, since near-term action would be expected to bring growth, inflation, and interest rates back to their long-term steady state overtime. As a result, efficient tracking of aggregate demand trends should also produce information value for yield curve trades. Specifically, strong macro demand should support curve-flattening positions and weak demand should support curve-steepening trades.

Quantamental indicators of aggregate demand growth

We test the theoretical relations between aggregate demand data and interest rate swap (IRS) returns using macro-quantamental data of the J.P. Morgan Macrosynergy Quantamental System (JPMaQS). Macro quantamental indicators capture the information state of the market with respect to macroeconomic activity. Unlike regular economic data, they are based solely on information that was available at the time of recording and are therefore suitable for backtesting trading ideas and implementing algorithmic strategies.

In JPMaQS there are four types of quantamental indicators that allow tracking various aspects and drivers of aggregate demand across 24 developed and emerging economies with liquid interest rate swap markets, which are Australia (AUD), Canada (CAD), Switzerland (CHF), Chile (CLP), Colombia (COP), Czech Republic (CZK), the euro area (EUR), the UK (GBP), Hungary (HUF), Indonesia (IDR), Israel (ILS), India (INR), Japan (JPY), South Korea (KRW), Mexico (MXN), Norway (NOK), New Zealand (NZD), Poland (PLN), Sweden (SEK), Thailand (THB), Turkey (TRY), Taiwan (TWD), the U.S. (USD), and South Africa (ZAR).

Excess estimated GDP growth trends: Fluctuations in GDP growth around long-term trends mostly reflect fluctuations in aggregate demand. We use two types of GDP growth trend (% over a year ago in 3-month moving averages) measures.

  • Technical growth estimates are real-time GDP growth trends based on vintages of standard econometric “nowcasting” estimates. Historic and current models are based on the simplest conventions and are recurrently reconstructed based on learning (view basic principles here). This process reduces the contamination of model hyperparameters by hindsight. View documentation here.
  • Intuitive growth estimates are real-time estimated recent GDP growth trends based on regressions that use the latest available national accounts data and monthly-frequency activity data, both in vintages form. The estimation relies on GLS regression with autocorrelated errors. Unlike standard academic models, the intention is to mimic the intuitive methods of market economists. View documentation here.

From these two types of growth trends, we subtract the concurrent information of the past 5 years’ median GDP growth rate. Thus, the indicator mainly represents deviations of economic activity and demand growth from what could be considered as “normal” at the time. Historically, quantamental measures of GDP growth trends have displayed pronounced swings during recession and recovery periods and only modest fluctuations in normal times.

Excess retail sales growth: Retail sales are just one sub-section of consumer spending but are well documented by timely data series in 23 of 24 currency areas with liquid IRS markets. India is the exception. We consider measures of excess retail spending growth as % over a year in 3-month moving averages or quarterly values, depending on the available frequency:

  • Excess nominal retail sales growth is the difference between nominal retail sales growth in local currency terms and the sum of medium-term GDP growth (5-year rolling median) and the official or implied inflation target. Not all countries release nominal retail sales growth. Chile, Indonesia, Israel, and Mexico only release retail volume data.
  • Excess real retail sales growth is the difference between real retail sales growth and medium-term GDP growth. Again, not all countries produce real retail sales statistics, but in these cases, JPMaQS approximates real values by using the CPI as the deflator.

View documentation for retail sales growth in quantamental format here. Historically, excess retail sales growth has been more volatile than excess GDP growth. Also, unlike excess GDP growth, excess nominal retail sales growth is driven not only by volume but also by price fluctuations.

Excess private credit growth: Private credit is not itself part of aggregate demand but it is related to it and indicative of the influence of interest rates on household and corporate spending. We consider two measures of excess private credit growth:

  • Simple excess private credit growth is the difference between (a) growth in private bank credit as % over a year ago and (b) the sum of medium-term real GDP growth and the inflation target.
  • Proportionate excess private credit growth is the difference between (a) the change in private bank credit as % of GDP over a year ago and (b) the sum of medium-term real GDP growth and the inflation target, times the share of private credit to GDP.

View documentation on private credit growth in quantamental format here. Private credit growth cycles have looked quite different from standard GDP growth cycles for most countries since 2000.

Excess merchandise import growth: Merchandise imports are the part of domestic aggregate demand that is met by goods produced in other currency areas. See the documentation of merchandise import trends in quantamental format here. Nominal local currency import growth is considered in two versions: % over a year ago in a 3-month moving average and % of the past 6 months over the previous six months, seasonally adjusted and annualized. As for retail sales and private credit, excess growth is estimated as the difference between outright growth and the sum of medium-term GDP growth (5-year rolling median) and the official or implied inflation target.

Excess merchandise import growth displays more cycles than real GDP and private credit growth. Fluctuations are driven by both price and volume effects.

Combining quantamental indicators to an aggregate demand score

This post means to deliver a proof of concept for the predictive power and economic trading value of macro demand-related quantamental indicators with respect to interest rate swap positions. For this purpose, we do not seek to optimize the use of these indicators but simply average over normalized versions and apply the averages as trading signals in all liquid markets. This process follows the principles of “risk parity” and “double diversification” that were explained in a previous post here. The normalization and aggregation are done in two steps.

First, we create individual (category) scores. This means that we normalize and average all versions of excess growth for each of the four economic categories: real GDP, retail sales, private credit, and merchandise imports. Normalization means that we sequentially estimate and divide by the standard deviation of the indicator category, using the data of the full panel up to the respective point in time. We do not subtract a mean, because all excess growth rates already have a natural neutral value of zero. Averaging means that we take the arithmetic mean of the concept sub-scores. If not all values are available for any combination of point-in-time and cross-section of an indicator, the score is calculated based on the available indicators alone.

The resulting demand-related scores for the four economic categories are visualized in the multi-line panel below. The individual scores are all positively correlated for the whole panel, but no pair has more than 50% correlation and the private credit score only exhibited 10-30% correlation with the other categories. Thus all subscores offer independent information value.

Second, we calculate a single aggregate demand score for each of the 24 currency areas. Normalization and aggregation follow the same principles as those applied to the individual category scores. The below panel shows the composite demand scores, which follow broadly similar larger cycles but with different exact forms and amplitudes and ample idiosyncratic mini-cycles.

Cross-market correlation of aggregate demand scores has been between near zero and almost 80%. The highest correlations can be seen between developed countries. The lower correlation across emerging markets illustrates the usefulness of these currency areas for a diversified rates strategy based on a single principle.

Aggregate demand score and duration returns

We first test the hypothesis that the point-in-time aggregate demand score, i.e., an information state of a broad estimate of excess demand, negatively predicts subsequent duration returns. The targets of the analysis are 5-year IRS fixed receiver returns across the 24 countries, vol-targeted at 10% annualized with monthly rebalancing in each market, and subject to market liquidity and tradability. The pattern and availability of these returns are shown in the line facet below.

We estimate the explanatory power of the aggregate demand score for subsequent IRS receiver returns from 2000 to 2023 (October) based on the Macrosynergy panel test (view post here). This adjusts targets and features of the predictive regression in a cross-country panel for common (global) influences. The stronger these global effects, the greater the weight of deviations from the period-mean in the regression, thus avoiding “pseudo-replication” without disregarding any country-specific experiences.

Panel tests show a negative and highly significant relation between demand scores as period-end information states and subsequent IRS receiver returns at monthly and quarterly frequencies. This negative relation seems to be stable, holding across various sub-periods.

Balanced monthly accuracy of return prediction, i.e., the average precision of positive and negative return prediction based on the composite demand score has been 52.3% across the panel since 2000. All sub-scores posted above-50% accuracy and balanced accuracy as well.

We estimate the economic value of a composite aggregate demand score by computing a naïve PnL according to standard rules used in previous posts. A naive PnL is calculated using regular rebalancing in accordance with the score at the end of each month for new positions at the beginning of the next month, allowing for a 1-day time-lapse for trading. The trading signals are limited to a maximum of 2 standard deviations as a reasonable risk limit. The naïve PnL does not consider transaction costs or compounding. For representation in charts, the PnL has been scaled to an annualized volatility of 10%

The simple naïve PnL according to these rules would have produced a long-term (2000 to October 2023) Sharpe ratio of 0.8 and a long-term Sortino ratio of 1.2, with almost no correlation with the U.S. treasury return. Value generation has been heavily seasonal and, for lack of proper risk management, concentrated on months of large signal values and market moves. However, good seasons in the demand-managed duration strategy have coincided with bad seasons for a long-duration portfolio, underscoring the complementarity of the macro demand-managed strategy.

Indeed, a demand-managed long-biased duration strategy would have produced roughly double the value of a simple risk parity long-only strategy. The below long-biased naïve PnL simply adds a value of 1 to the normalized composite demand score. The long-biased strategy would have produced a long-term Sharpe ratio of 0.9 and a Sortino ratio of 1.3, compared to 0.4 and 0.6 for the long-only strategy.

Each sub-score of the aggregate demand composite would have produced positive trading value without long bias or outperformed the long-only duration portfolio with long bias. The individual long-term Sharpe ratios for 2000 to October 2023 were 0.6 (excess GDP growth), 1.0 (excess retail sales growth), 1.0 (excess private credit growth), and 0.7 (excess merchandise import growth).

Aggregate demand score and curve flattening returns

The second hypothesis to be tested is that the point-in-time aggregate demand score positively predicts the returns on a curve-flattening trade. As argued above excess demand supports a temporarily tighter monetary stance, i.e., higher rates in the near term with a lesser impact on medium-term yields. The target returns of the analysis are 5-year IRS fixed receivers versus 2-year IRS payers, with both legs of the trade vol-targeted at 10% annualized return variation to remove systematic directionality of the trade.

Indeed, the Macrosynergy panel test shows a clear and significant positive correlation between end-of-period aggregate demand scores and subsequent monthly or quarterly curve flattening returns for the full panel of markets. Balanced accuracy of monthly return predictions has been higher than for directional IRS return predictions at 53.7%. All sub-scores posted above 50% accuracy and balanced accuracy as well.

To assess the economic value of the macro demand score for curve trades we calculate a naïve PnL according to the same rules that were applied to the directional IRS receiver positions.

PnL generation of the demand score has been strong and consistent across time. The 24-year Sharpe ratio has been 1.2 and the Sortino ratio 1.9, with almost no correlation of the PnL with the 10-year U.S. treasury return. Unlike for the directional strategy, the curve strategy posts higher performance ratios for the composite macro demand signal than for any of the sub-scores. The Sharpe ratios of the sub-scores for 2000 to October 2023 were 0.9 (excess GDP growth), 0.6 (excess retail sales growth), 0.4 (excess private credit growth), and 1.1 (excess merchandise import growth).

Like the directional strategy, the curve strategy posted its higher returns in the 2020s than in the 2000s and 2010s. This is a reminder of the natural seasonality of demand-based strategies. They typically produce their strongest signals in times of large deviations of demand from long-term trends, i.e., in the context of business cycle swings and credit booms and busts. The 2020-2023 produced large negative and positive signals within a fairly short span of time. By contrast, the period 2012-2018 only produced weak signals for the majority of currency areas.

Ralph Sueppel is managing director for research and trading strategies at Macrosynergy. He has worked in economics and finance since the early 1990s for investment banks, the European Central Bank, and leading hedge funds.