Sector ETF Analysis: Relative Performance And Flows

By Shane Obata  @sobata416

Introduction

The purpose of this analysis was to examine the relationships between relative performance and ETF flows. Relative performance was defined as the excess returns of nine sector SPDR ETFs over the S&P 500. ETF flows were defined as the net weekly flows, adjusted for price movements.

From a practical standpoint, it was useful to examine relative performance because adept sector rotation can lead to alpha generation. For example, a portfolio manager (PM) who was overweight financials and underweight energy from 2009 to 2017 would have fared better than another PM whose allocations were in line with the S&P 500’s sector weights.

It was also useful to examine flow data because they often reflect sentiment. For example, net inflows into a top performing sector may indicate irrational exuberance and or performance-chasing.

This analysis was novel in that it considered relative performance instead of absolute performance. Moreover, it addressed some of the limitations of past research, which are covered in the Literature Review section of this paper. For example, past research focused on monthly and daily flows. Both of these frequencies may be suboptimal since the former are too sparse and the latter are too noisy. This analysis focused on weekly flows.

The results of the analysis were quite good. On average, the core models exhibited higher explanatory power than ones presented in past research. In addition, various market-based variables were considered. Some of them were highly significant and helped to add explanatory power to the core models.

Literary Review

“Fund Flows and Underlying Returns: The case of ETFs” – Arsenio Staer – Q1, 2017

• Model:

• This paper looked at the effects of flows (X) on returns (Y)
• ETF assets grew at >27% per annum since their introduction in 1993
• ETFs exhibited a statistically significant and cross-sectionally consistent positive association with contemporaneous underlying index returns
• Daily data showed contrarianism and monthly data displayed performance chasing
• A 1SD positive flow shock corresponded to a positive impulse response in returns, followed by a meaningful reversal by t=5

• A 1SD positive flow shock corresponded to a small positive impulse in flows, followed by a complete reversal by t=5
• A 1SD positive return shock corresponded to a negative impulse response in flows, followed a stabilization period from t=5 onwards
• A 1SD positive return shock corresponded to a negative impulse response in returns, followed a stabilization period from t=5 onwards

Conclusions:

– ETFs exhibited a significant positive relationship with contemporaneous index returns
– There is substantial evidence of price reversals following shocks in flows

“Impact of rising interest rates on equity markets” – TIAA-CREF Asset Management – Q3, 2015

• The last 12 rate tightening cycles led to mixed results for equities, which gained in half of the cycles
• Interest rates were not enough. Core inflation levels and other variables were also key
• There have been 8 cycles since 1972. Sub-groups that performed well more than 50% of the time were Non-US Developed Markets (n=8) and Emerging Markets (n=4)
• REITs and Dividends also performed well (n=3)
• Sector performance: Financials, particularly banks, and utilities tended to underperform during tightening cycles. Industrials tended to outperform
• In this cycle, a period of sustained low rates has contributed to strong performance in “higher-yielding equity assets” such as Dividend Aristocrats, Utilities, REITs, MLPs, etc.

• QE in Europe etc. has been holding down US rates, which have offered relative value vs. counterparts
• Below “normal” rates should continue to artificially inflate demand for dividend paying stocks and US sovereign debt

Conclusions

– Rate tightening cycles yielded mixed results for equities. Important to consider other variables such as inflation
– Low interest rates have increased the demand for high-yield equity products

“What drives ETF flows?” – Clifford et al. – Q3, 2013

• Model:

• This paper looked at the effects of returns (X) on flows (Y)
• ETF flow chased past performance and this return chasing was particularly strong in broad market index funds
• Flows did not correctly anticipate returns

Conclusions

– Flows chased returns in ETFs
– There was no convincing evidence “that flows are able to successfully anticipate future performance

“Movements in the Term Structure of Interest Rates” – Robert B. Bliss – Q4, 1997

• Movements in stock prices are largely idiosyncratic. In other words, it is hard to hedge one with another security with another
• It is also hard to hedge stock price movements with index futures and other derivatives
• Movements in bond prices are largely systemic. They are easy to hedge with a small number of bonds of different maturities or with rates derivatives
• Explanatory models were more useful for interest rate changes than stock price changes. This speaks to the complexities of idiosyncratic risks
• Using portfolios instead of individual securities helped to increase the explanatory power of factor based stock models
• Factor 1 – a proxy for the level of rates – explained most of the variance in returns. Adding Factor 2 – the slope – also helped. Adding Factor 3 – the curve- added marginal explanatory power

Conclusions

– It is difficult to hedge against movements in stock prices due to idiosyncratic risks
– It is easy to hedge against movements in interest rates since movements in bond prices are largely systemic

“Common Factors Affecting Bond Returns” – Robert Litterman and Jose Scheinkman – Q2, 1991

• Most of the variation in returns on all fixed-income securities can be explained in terms of 3 factors: Level, Steepness and Curve

• When rates are high, returns on long bonds will be closer to returns on short ones
• When rates are low, returns on long bonds will be farther away from returns on short ones
• For duration hedged portfolios, changes in values were attributable to changes in spreads. In other words, duration hedging eliminated systemic risk
• “By increasing the number of instruments within the same maturity spectrum, we can make the dollar share of each security arbitrarily small…and hence make the variance of the return per dollar invested in the portfolio as small as we want…”
• Factor 1 – a proxy for the level of rates – explained 89.5% of the variance in returns over the Risk Free Rate. Factor 2 explained 81% of the remaining variation
• Curvature was highly correlated with implied volatility
• Duration hedging greatly reduced dollar exposure to the level and steepness of rates (when considered in the hedging process). Remaining variance was largely explained by curvature

Conclusions

– The level and steepness of rates explained the vast majority of the variance in returns
– Hedging with a portfolio of bonds was more effective than hedging with one bond

Analysis

Legend:

Inputs

Formulas

Data From Another Sheet

Data From Another Workbook

*Pricing data was obtained from Yahoo Finance, the St. Louis Fed, the CBOE and Investing.com. Flow data was obtained from Bloomberg. All data cover the March 6^th, 2009 to February 10^th, 2017 timeframe.

Absolute Performance

For this analysis, the nine SPDR sector ETFs were divided into three groups: Bond Proxies, Inflation Proxies and Other Sectors. The Bond proxies were the sectors that were most affected by interest rates, the Inflation Proxies were the sectors that were most affected by inflation and the Other Sectors were affected by other variables.

Bond Proxies performed well over the timeframe. Both Staples (+238%) and Utilities (+190%) exhibited sizeable gains…

..This was not surprising because interest rates, especially at the long end, were in a downtrend from Q3’10 to Q3’17.

Inflation Proxies also performed well. Discretionary (+495%) led the group, followed by Financials (+451%), Industrials (+396%), Healthcare (+272%) and Materials (+240%)…

..Inflation, represented by 10-Year Breakevens, was rising at a steady pace over the time frame. This helped to lift the Inflation Proxies, which are mostly Cyclical in nature (excluding Healthcare).

Other Sectors showed mixed results. Technology (+331%) outperformed Energy (+123%). The latter was rising on a similar trajectory until oil prices started to collapse in H2’14.

Relative Performance

The Bond Proxies performed well in absolute terms but not in relative terms. Both Utilities and Staples lagged the S&P 500 over the timeframe…

..For the Inflation Proxies, relative performance was mixed. Discretionary, Financials & Industrials outperformed the SPX while Healthcare and Materials underperformed…

..The Other Sectors also showed mixed results. Technology outperformed while Energy underperformed by the widest margin of all the sectors.

In terms of Summary Statistics, Discretionary was the leader while Energy was the laggard. The former displayed one of the highest Minimum returns, the highest Average & Median Returns and the lowest Standard Deviation. The latter showed the lowest Minimum return, Average & Median Returns and the third highest Standard Deviation.

The correlation matrix yielded some interesting results. The Bond Proxies displayed strong positive correlations while the Inflation Proxies showed mixed results. Cyclical sectors such as Financials and Industrials showed moderate negative correlations with Defensive sectors such as Utilities, Staples and Healthcare.

Flows

The Bond Proxies saw inflows during the European debt crisis in 2011, outflows during the Taper Tantrum in 2013, inflows during the collapse of oil in H2’14 and outflows during the Trump reflation trade in H2’16…

..The Inflation Proxies, with the exception of Materials, saw big inflows over the timeframe. The spikes in Financials, Healthcare and Industrials that occurred in Q4’16 are attributable to the Trump reflation trade…

..For the Other Sectors, the data were counterintuitive. Energy saw sustained buying pressure from H1’14 to Q1’17, despite the collapse in oil prices. On the other hand, cumulative flows into Technology peaked in H1’14.

In terms of Summary Statistics, Discretionary was a leader while Materials was a laggard. The former displayed the highest Average & Median flows and the third highest Max flows while the latter showed the highest Standard Deviation.

The correlation matrix was not remarkable, with low to moderate relationships across the board. That said, it was interesting to note the weak negative correlation between Defensive Utilities and Cyclical Discretionary, Industrials and Financials.

Relative Performance vs. Flows

There were a few notable differences between the Relative Performance and Flows data. Healthcare, Materials and Energy underperformed the S&P 500. Even so, each of those sectors saw massive inflows over the timeframe. In contrast, Financials was the second best performing sector relative to the S&P 500 but was ranked sixth in terms of flows.

The correlation matrix showed that there was a consistent, positive relationship between relative performance and flows.

Regressions – Relative Performance

The core models for Relative Performance were defined as follows:

Relative Performance = Intercept + Cumulative Flows + CF – Lag 1 + CF – Lag 2 + RP – Lag 1 + RP – Lag 2 + 2s10s + 10-Year Yield + 10-Year Breakeven

The first thing to note from the regression outputs is that Cumulative Flows were highly significant for each of the nine sector ETFs. All of the t-stats were positive, which indicates that net inflows were associated with relative outperformance.

The 10-year breakeven inflation rate and emerging market (EM) equities were also very significant. The t-stats for inflation were positive for Cyclical sectors such as Materials, Discretionary, Industrials and Financials and negative for Defensive sectors such as Utilities, Staples and Healthcare. The t-stats for EM equities were positive for Energy and Materials. This is due to the fact that many emerging markets are commodity-based. The t-stats for EM equities were negative for Defensive Utilities and Staples, which tend to underperform when investors are risk seeking.

The other variables showed mixed significance across sectors.

In terms of explanatory power, the best models were for Staples (49.58% R²) and Utilities (43.91% R²). The worst ones were for Technology (7.41% R²) and Discretionary (13.32% R²). In aggregate, the models had an average R² of 28.36%, which is decent considering how much idiosyncratic risk is embedded in equities.

The table for incremental R² shows how much explanatory power was added at each step. As noted above, Cumulative Flows were important for most sectors.

Aside from that, there were three main observations.

• The slope (2s10s) and level (10-Year) of interest rates were essential for the Bond Proxies’ models. Those two factors added 22.53% and 22.67% of explanatory power for Utilities and Staples, respectively
• Inflation was a key contributor for many sectors
• With the exception of EM equities, most of the independent variables outside of the core models did not add much incremental value

Regressions – Cumulative Flows

The core models for Cumulative Flows were defined as follows:

Cumulative Flows = Intercept + Relative Performance + RP – Lag 1 + RP – Lag 2 + CF – Lag 1 + CF – Lag 2 + 2s10s + 10-Year Yield + 10-Year Breakeven

Lags were more important in the flow models than they were in the relative performance ones. For example, lag one for cumulative flows was significant for six of the nine sector ETFs. In each of the six cases, the t-stats were negative. This suggests that inflows flows do not beget more inflows.

The other variables showed mixed significance across sectors.

In terms of explanatory power, the flow models were weaker than the relative performance ones. The best models were for Industrials (36.02% R²) and Financials (28.95% R²). The worst ones were for Technology (6.99% R²) and Healthcare (12.99% R²). In aggregate, the models had an average R² of 19.16%.

As noted above, Relative Performance was important for most sectors.

Aside from that, there were three main observations.

• The lags for Cumulative Flows were meaningful for a number of sectors such as Energy, Materials, Staples and Industrials
• The slope (2s10s) and level (10-Year) of interest rates were essential for several models; however, they did not add much power for the Bond Proxies’ ones
• Most of the independent variables outside of the core models did not add much incremental value. After rates (bucket 4), none of the variables added more than 1.00% of incremental R² on average

Conclusion

Using weekly data, I found that Cumulative Flows exhibited a consistent and significant positive association with contemporaneous Relative Performance. The relationship was equally as strong in the reverse direction; both models showed an average R² of 8.30% after the first bucket.

In addition, I found evidence that the level & slope were very important for the Bond Proxies and that inflation was key for both Cyclical (positive association) and Defensive sectors (negative association).

Outside of the core models, the significance of variables differed across sectors. That said, those variables did not add much incremental explanatory power.

Moreover, the lagged flow data displayed contrarian and not performance-chasing behavior.

These findings provide a novel contribution to financial literature by examining relative performance instead of absolute performance and by showing that weekly data may be more useful than daily or monthly data.

Appendix

XLE – Relative Performance

XLE – Cumulative Flows

XLU – Relative Performance

XLU – Cumulative Flows

XLK – Relative Performance

XLK – Cumulative Flows

XLB – Relative Performance

XLB – Cumulative Flows

XLP – Relative Performance

XLP – Cumulative Flows

XLY – Relative Performance

XLY – Cumulative Flows

XLI – Relative Performance

XLI – Cumulative Flows

XLV – Relative Performance

XLV – Cumulative Flows

XLF – Relative Performance

XLF – Cumulative Flows

Article by Shane Obata