Trend-following strategies rely on the persistence of market trends. Such persistence can arise from the gradual dissemination of information or behavioural biases. In light of these inefficiencies, trends that coincide with supporting economic information (macro tailwinds) are more likely to persist than those accompanied by opposing macro information (macro headwinds). As a result, a macro-enhancement of standard trend-following signals should produce better investment returns.
This post supports this proposition for the U.S. Treasury market over the past 32 years. It tests simple macro enhancement types for directional return trends and curve-flattening return trends. In all cases, macro enhancement would have materially improved predictive power and backtested trading profits. This echoes previous research for other asset classes that illustrated the complementarity of price and economic information in systematic trading.
Please quote as “Regis, Glenn and Sueppel, Ralph, ‘U.S. Treasuries and macro-enhanced trend following,’ Macrosynergy research post, December 2024.”
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. Everyone with DataQuery access can download data excluding the latest months. Moreover, J.P. Morgan offers free trials on the full dataset for institutional clients. An academic research support program sponsors data sets for relevant projects.
This post ties in with this site’s summary of macro trends
Market trends and the importance of macro tailwinds and headwinds
Trend following refers to trading strategies that seek profit by identifying persistent market trends and setting up positions following them. Trend persistence may partly reflect the gradual dissemination of information. However, it can reflect inefficient or excessive flows:
- The “disposition effect” is a well-documented behavioural bias that leads investors to liquidate profitable positions and hang on to loss-making trades (view post here). It delays market adjustments to changing conditions. The bias can be motivated by regret avoidance (a tendency to minimize the risk of future emotional discomfort) and loss aversion (a preference for avoiding losses over acquiring equivalent gains).
- “Herding” is another well-researched phenomenon and denotes a deliberate decision to imitate the actions of others rather than relying on one’s own analysis. It can be motivated by believing in the “wisdom of crowds” (superiority of decentralised collective judgment) or psychological comfort. Herding can be efficient for the individual, particularly in markets with ample private and “inside” information, but increases the chances of inefficient flows in the aggregate (view post here).
- “Lazy trading” refers to a staggered response of market participants to changing conditions. Private investors, in particular, only re-allocate their portfolios infrequently, and even institutional investors often follow lengthy due processes (view post here). Some types of trend-following themselves can be viewed as an institutionalised form of lazy trading, particularly if they rigidly cling to long-term trend metrics.
Trends in financial market prices change financial and economic conditions. For example, rising equity prices reduce the cost of raising capital and increase available collateral for borrowing. Real currency appreciation increases local purchasing power, but it also reduces the cost competitiveness of local producers. Declining interest rates reduce borrowing costs relative to income, and if they reflect lower and well-anchored inflation expectations, they reduce term premia and encourage both lending and borrowing.
All this accords with mainstream macroeconomic theory, which proposes that asset prices are part of a general economic equilibrium. This implies that to the extent that financial markets are inefficient, price trends can fall short of what is required to re-balance the economy after a change in conditions or go too far. Macroeconomic information can help tell the former from the latter:
- If economic information states show price pressure in the direction of the prevailing market trend, they are macro tailwinds for that trend. If market trends mainly reflect the gradual dissemination of information, macro tailwinds should be prevalent. For example, if there is a downward drift in bond yields while credit growth and aggregate demand in the economy are still soft, the economic developments support the continuation of the bond market trend.
- If economic trends push prices in the opposite direction of the prevailing market trend, they are macro headwinds. To the extent that behavioural biases, herding, and institutional inertia exaggerate price trends, macroeconomic information should turn into headwinds after some time lag. For example, if a downward trend in bond yields coincides with evidence of strong credit and demand growth, the economic information becomes a headwind for the prevailing fixed-income market trend.
This is important because, assuming partial information inefficiency, price trends with macro tailwinds are more likely to persist than those with macro headwinds. This is the main proposition to be tested in this post: macro-enhanced market trends are superior trend-following signals.
Macro tailwinds and headwinds in high-grade government bond markets
Government bond trends are one of the most important drivers of financial conditions. In developed financial markets with high-grade sovereign debt, government bond yields serve as a benchmark for private credit rates and determine the value of government debt as collateral for secured lending. Bond yields also indicate long-term interest rate expectations, inflation expectations, and risks around them (view post here).
We propose to track macroeconomic developments that are associated with interest rates. To test the value of such trackers based on history, we must use meaningful point-in-time macro information states, also known as macro-quantamental indicators. These can be downloaded from the J.P. Morgan Macrosynergy Quantamental System (“JPMaQS”) for a broad range of currency areas. In particular, we focus on six types of macro factors, all of which should put downward pressure on future monetary policy rates and hence support positive bond returns:
- Inflation shortfall: This is the negative of excess inflation rates versus countries’ effective inflation target. Sub-target inflation mandates central banks to run accommodative policies and err on the side of monetary easing in times of uncertainty. The inflation rates used here are headline CPI inflation, % over a year ago (documentation here), various core CPI/PCE inflation rates, i.e., % over a year ago, % 6 months over 6 months, seasonally and jump-adjusted annualised rate, and % 3 months over 3 months, seasonally and jump-adjusted annualised rate (documentation here), excess 1-year ahead CPI inflation expectation (documentation here), and wage growth, % over a year ago, 3-month moving average or quarterly (documentation here).
- Money and credit growth shortfall: This is the negative of money and credit growth rates relative to the medium-term nominal GDP trend. All other things equal, a slump in money and credit calls for a more accommodative central bank policy. The constituents of this factor are broad money growth (documentation here), narrow money growth (documentation here), and private credit growth (documentation here), all as % over a year ago, as well as private credit growth as difference over a year ago in % of GDP (documentation here).
- Labor market slack: Soft labor market conditions can be approximated by trends in employment and unemployment. Labor market slack or slackening means downside risk for economic, wage, and income growth and warrants easier monetary policy. We approximate slackness in the labor market by the negative of employment growth, % over a year ago, 3-month average (documentation here) versus medium-term workforce growth, the change in the unemployment rate, difference over a year ago, 3-month average (documentation here), and level of the unemployment rate, 3-month average, over and above its 5-year median (documentation here).
- Growth and demand shortfall: This is the negative of excess growth in real GDP and excess nominal and real domestic demand. Weak activity and demand growth point to downside risks for the economy and encourage monetary policy to be more supportive. The excess GDP growth rates are “intuitive” nowcasts of GDP growth (documentation here) and “technical” nowcasts of GDP growth (documentation here), % over a year ago, 3-month averages versus historic 5-year averages. The demand indicators are nominal retail sales growth (documentation here), nominal merchandise import growth (documentation here), and real private consumption growth (documentation here), all % over a year ago as 3-month moving averages versus the 5-year average of either nominal or real GDP growth.
- Economic sentiment shortfall: This factor is based on business and consumer confidence scores with negative signs. Low sentiment points to downside risk for economic activity and skews the monetary policy bias to the accommodative side. The factor uses the negatives of consumer sentiment scores for manufacturing (documentation here), services (documentation here), construction (documentation here), and consumers (documentation here), all as 3-month moving averages.
- Real estate shortfall: This is the negative of excess residential real estate price growth. Underperforming real estate prices point to financial stability risk and encourage monetary policy accommodation. The basis for the factor is both nominal and real house price growth (documentation here), as % over a year ago, relative to medium-term nominal and real GDP growth, respectively.
For the empirical analysis, we focus on the U.S. Treasuries and, for now, disregard other bond markets. The U.S. bond market dominates international bond markets and has the longest history of macro-factor data and target returns, going back to 1992. However, we use data panels for all developed markets as the basis for normalising the macro indicators of the U.S. Meanwhile, the targets of analysis are (1) generic 5-year U.S. Treasury bond excess returns (documentation here) and (2) Treasury curve-flatting returns, which are approximated as the difference between the return on vol-targeted position in the 10-year Treasury and a vol-targeted position in the 2-year Treasury (documentation here).
We derive normalised scores for all the above macro-quantamental categories and the composite factors. This means that we divide the factor constituents, which all have theoretical neutral levels at zero, by the estimated standard deviations of the panel of developed currency areas based on information up to real-time date. We also winsorise (cap or floor) the scores at three standard deviations to de-emphasise the effects of outliers. Then, we take a weighted average of all constituents and re-normalise to obtain factor scores. The graph below shows that cross-factor correlation has been mostly positive and particularly strong across factors related to the business cycle and between credit and real estate price growth.
Finally, we take an unweighted average of the conceptual macro factor scores and re-normalize it to obtain a “composite macro support score” for directional Treasury market trends. Given the nature of the macro factors, it may also be called an “expected monetary policy accommodation score” because all macro factors are presumed to bias central bank actions and talk to the accommodative side. For this reason, the negative of the score can serve as a macro support score of curve flattening trends. In a credible monetary regime, expectations of monetary tightening affect the short end of the curve more than the long end. Distant forward rates are more anchored by the inflation target and estimated natural real interest rates.
The graph below shows the evolution of the composite macro support score for U.S. Treasury returns since 1992. The score has mostly been negative, i.e., opposing positive return trends and supporting negative trends. This is plausible because yields have mostly declined over the past three decades, supporting growth and the expansion of financial leverage. Economic developments have tempered this downward drift. However, macro factors have supported positive Treasury return trends during financial crises and recessions. Finally, the economic rebound and rise in inflation after the COVID crisis produced a particularly unfavourable macro environment for Treasuries.
Not all of the macro indicators that are used in this score have empirically added value for Treasury trend adjustment. This score is not optimized. Thus, headline inflation failed to predict directional Treasury returns, and credit and real estate price growth did little to predict curve-flattering returns. However, all indicators have theoretically plausible predictive power under some conditions, thus justifying their tracking for future purposes. Also, keeping historical “failures” in the score produces more realistic backtests.
Trend adjustment for directional U.S. Treasury positions
We defined a standard return trend based on excess returns, a 20-day moving average minus a 100-day moving average of the 5-year Treasury excess return index. Excess returns mean treasury returns net of funding costs. The choice of average periods was guided by plausibility for a medium-term trend in a highly liquid market but is not optimised. The point of this post is to show the benefits of macro enhancement, not to assess the merit of particular trend identification methods.
The chosen directional market return trend would have been a modestly positive predictor for subsequent monthly Treasury returns, albeit with a probability of significance of only 70%, below standard thresholds. As shown further below, related trend-following strategies have nevertheless produced positive PnLs due mainly to their long bias and positive carry.
The composite macro support score also posted positive predictive power with respect to 5-year Treasure excess returns but with greater correlation and high significance. The profitability of such a score as a signal would have been modest, however, due to its strong short bias.
Market return trends and macro trends have naturally positively correlated, reflecting common underlying developments, such as disinflation or the state of the business cycle. However, their concurrent correlation has been well below 50%, illustrating that the two concepts offer diversified guidance on market direction. Both types of trends are valid signals, but they have complementary strengths. The market trend signal is very timely but not specific in its message; it could be driven by economic changes or inefficient flows. The macro trend factors convey information with (publication-related) delays but are very specific in their meaning. Complementarity argues for joint application.
There are two principal simple types of macro enhancement for directional signals:
Modified trends reduce or enhance the original return trend by multiplying it with a sigmoid function of the macroeconomic factor score.
The adjustment function has the following form:
modified_trend = ((1 – sign(trend)) + sign(trend) * coef) * trend
for
coef = 2/(1 + exp(-3 * macro_score))
where trend means the original trend, sign() returns the sign of its element as 1/-1, and macro_score denotes the composite macro score. The translation of the macro score into the adjustment coefficient is visualized below.
If the macro score points in the same direction as the market trend, the latter is multiplied by a value above one up to a maximum of 2. If the macro score points opposite the market trend, the sigmoid function’s value is below one and can reach a minimum of zero. Hence, trend modification only changes the strength of the original trend signal, never its direction.
The chart below shows the simple and modified directional return trends for the 5-year U.S. treasury since 1992. The modified signal either contains or increases the simple trend signal both in negative and positive territory and results in “fatter tails” of the distribution.
Balanced trends are simple averages of normalized return trends and macro scores. This means that if the macro score points to the opposite direction of the market trend and is larger, it can override the market trend. Balanced trend signals give equal importance to macro and market information and can alter the character of the original trend signal more strongly.
In the case of the U.S. treasury market, balanced trend signals have changed the character of the strategy modestly. The modest impact of balancing reflects that return trends and macro scores have been positively correlated.
The scatterplots, regression lines, and significance statistics below illustrate how modification and balancing affect the predictive power of the directional market trend signal at a monthly frequency. We present regressions separately for both halves of the sample period to check for stability of the relationship over time. Both modification and balancing clearly improve the predictive power of the macro trend for all subsamples. The macro score alone would have delivered a sizable monthly correlation on its own with a high probability of significance.
However, macro enhancement does not increase signal accuracy, i.e., the share of correctly predicted return directions and balanced accuracy, which is the average of correctly predicted positive and negative returns. This mainly reflects the short bias of macro factors, with 65% of monthly signals being negative. This may point to difficulties in setting correct neutral levels for macro factors.
To assess the economic value of the macro-enhancement of the directional Treasury market trend, we calculate simplistic naïve PnLs based on simple, modified, and balanced trend signals. PnL estimation follows standard rules used in many Macrosynergy research posts. Positions are taken based on normalised simple or adjusted trends at the end of the month and held over the next month, with a one-day slippage added for trading. The PnL does not consider transaction costs and risk management rules because both are specific to the trading institution and the portfolio size. The long-term volatility of the PnL for positions across all currency areas has been set to 10% annualized for comparability in charts.
Both modification and balancing would have increased risk-adjusted returns of a simple market trend signal from very low levels. Simple trend following in the Treasuries without enhancement has produced a very modest long-term Sharpe ratio of 0.2 and a Sortino of 0.3, with just below 20% correlation to the market. The low values are plausible since we apply a simple and “cheap” signal to one single contract in a highly efficient market. Macro modification of trends would have raised Sharpe and Sortino to 0.4 and 0.5, respectively; balancing would have achieved Sharpe and Sortino ratios of 0.5 and 0.7.
Trend adjustment for U.S. Treasury curve positions
We extend the concept of return trends to 2-10s flatting positions in the U.S. treasury market, whereby both the short position (in the 2-year bond) and long position (in the 10-year bond) are volatility targeted, based on past daily returns. The trend is defined as the difference between a 50-day and 200-day moving average of a flatting position return index. We chose a longer lookback than for the directional trend, given that transaction costs are higher, and the information efficiency of that market segment is presumed to be lower.
The curve flattening trends can be modified and balanced similarly to the directional trends, with the only exception that the sign of the composite macro factor is now negative. Since the factor measures the support for monetary easing, high values argue for disproportionate vol-targeted gains in the 2-year segment since it is more affected by the immediate policy rate outlook. This means the composite macro support factor favours curve steepening positions.
The below graphs illustrate the impact of modification and balancing for flattening trends since 1992. Neither adjustment changes the characteristics of the signal dramatically, but the differences are still clearly visible.
As for directional trends, the macro enhancement of curve-flattening trends increases their predictive powers through modification and balancing for all sub-samples in the below analysis.
In contrast with directional trend signals, macro-enhancement of curve trends would have also increased the accuracy and balanced accuracy of signals. While the directional macro trend had a strong short bias, the curve macro trend had a strong flattening bias.
As a result, macro-enhancement of curve flattening trends has produced notable improvements of risk adjusted returns. The Sharpe and Sortino ratios of the simple trend following the signal would have been 0.4 and 0.5, respectively. Both the modified and balanced trend signals would have lifted these to 0.6 and 0.8, respectively.
