Technical Trading

Are technical rules applied to pairs trading attractive after correcting for data snooping bias? In their March 2018 paper entitled “Pairs Trading, Technical Analysis and Data Snooping: Mean Reversion vs Momentum”, Ioannis Psaradellis, Jason Laws, Athanasios Pantelous and Georgios Sermpinis test a variety of technical trading rules for long-short trading of 15 commodity futures, equity indexes and currency pairs (all versus the U.S. dollar) frequently used on trading websites or offered by financial market firms. Specifically, they test 18,412 trend-following/momentum and contrarian/mean-reversion rules often applied by traders to past daily pair return spreads. They consider average excess (relative to short-term interest rate) return and Sharpe ratio as key metrics for rule selection and performance measurement. They use False Discovery Rate (FDR) to control for data snooping bias, such that 90% of the equally weighted best rules in FDR-corrected portfolios significantly outperform the benchmark. Most tests are in-sample. To test robustness of findings, they: (1) account for one-way trading frictions ranging from 0.02% to 0.05% across assets; (2) consider five subperiods to test consistency over time; and, (3) perform out-of-sample tests using the first part of each subperiod to select the best rules and roughly the last year to measure performance of these rules out-of-sample. Using daily prices of specified assets and daily short-term interest rates for selected currencies during January 1990 (except ethanol starts late March 2006) through mid-December 2016, they find that:

Keep Reading

Are bond market investors generally shrewder than their stock market counterparts, such that bond yield tops (bottoms) anticipate stock market bottoms (tops)? To investigate, we employ both a monthly lead-lag analysis and a comparison of bond yield and stock market tops and bottoms. We define “top” and “bottom” as the highest (lowest) value in a rolling window that extends from 30 months in the past to 30 months in the future (a total window of five years). Using monthly levels of Moody’s yield on seasoned Aaa corporate bonds and the Dow Jones Industrial Average (DJIA) during October 1928 through February 2018 (about 90 years) and monthly levels of the 10-year government bond interest rate and the stock market from Robert Shiller during January 1871 through February 2018 (about 148 years), we find that: Keep Reading

Can advanced regression techniques identify monthly cross-industry lead-lag return relationships that usefully indicate an industry rotation strategy? In their January 2018 paper entitled “Dynamic Return Dependencies Across Industries: A Machine Learning Approach”, David Rapach, Jack Strauss, Jun Tu and Guofu Zhou examine dynamic relationships between past and future returns (lead-lag) across 30 U.S. industries. To guard against overfitting the data, they employ a machine learning regression approach that combines a least absolute shrinkage and selection operator (LASSO) and ordinary least squares (OLS). Their approach allows each industry’s return to respond to lagged returns of all 30 industries. They assess economic value of findings via a long-short industry rotation hedge portfolio that is each month long (short) the fifth, or quintile, of industries with the highest (lowest) predicted returns for the next month based on inception-to-date monthly calculations. They consider three benchmark hedge portfolios based on: (1) historical past average returns of the industries; (2) an OLS-only approach; and, (3) a cross-sectional, or relative, momentum approach that is each month long (short) the quintile of industries with the highest (lowest) returns over the past 12 months. Using monthly returns  for 30 value-weighted U.S. industry groups during 1960 through 2016, they find that:

Keep Reading

A subscriber proposed: “It would be interesting to do an analysis of the Chemical Activity Barometer [CAB] to see if it has predictive value for the stock market. Either [look] at stock prices when [CAB makes] a two percent pivot down [from a preceding 6-month high] as a sell signal and one percent pivot up as a buy signal…[or when CAB falls] below its x month moving average.” The American Chemistry Council claims that CAB “determines turning points and likely future trends of the wider U.S. economy” and leads other commonly used economic indicators. To investigate its usefulness for U.S. stock market timing, we consider the two proposed strategies, plus two benchmarks, as follows:

1. CAB SMAx Timing – hold stocks (the risk-free asset) when monthly CAB is above (below) its simple moving average (SMA). We consider SMA measurement intervals ranging from two months (SMA2) to 12 months (SMA12).
2. CAB Pivot Timing – hold stocks (the risk-free asset) when monthly CAB most recently crosses 1% above (2% below) its maximum value over the preceding six months. We look at a few alternative pivot thresholds.
3. Buy and Hold (B&H) – buy and hold the S&P Composite Index.
4. Index SMA10 – hold stocks (the risk-free asset) when the S&P Composite Index is above (below) its 10-month SMA (SMA10), assuming signal execution the last month of the SMA measurement interval.

Since CAB data extends back to 1912, we use Robert Shiller’s S&P Composite Index to represent the U.S. stock market. For the risk-free rate, we use the 3-month U.S. Treasury bill (T-bill) yield since 1934. Prior to 1934, we use Shiller’s long interest rate minus 1.59% (the average 10-year term premium since 1934). We assume a constant 0.25% friction for switching between stocks and T-bills as signaled. We focus on number of switches, compound annual growth rate (CAGR) and maximum drawdown (MaxDD) as key performance metrics. Using monthly data for CAB, the S&P Composite Stock Index, estimated dividends for the stocks in this index (for calculation of total returns) and estimated long interest rate during January 1912 through December 2017 (about 106 years), and the monthly T-bill yield since January 1934, we find that: Keep Reading

In reaction to “Combine Market Trend and Economic Trend Signals?”, a subscriber suggested adding an interest rate trend signal to those for the U.S. stock market and U.S. unemployment rate for the purpose of timing the S&P 500 Index (SP500). To investigate, we look at combining:

• The 10-month simple moving average (SMA10) for the SP500. The trend is positive (negative) when the index is above (below) its SMA10.
• The 12-month simple moving average (SMA12) for the seasonally adjusted U.S. civilian unemployment rate (UR). The trend is positive (negative) when UR is below (above) its SMA12.
• SMA12 for the average monthly 3-month U.S. Treasury bill yield (T-bill). The trend is positive (negative) when T-bill is below (above) its SMA12.

We consider scenarios when the SP500 trend is positive, the UR trend is positive, the T-bill trend is positive, at least one trend is positive (>=1), at least two trends are positive (>=2) or all three trends are positive (All). For total return calculations, we adjust the SP500 monthly with estimated dividends from the Shiller dataset. When not in the index, we assume return on cash from the broker is the specified T-bill yield. We focus on gross compound annual growth rate (CAGR), maximum drawdown (MaxDD) and annual Sharpe ratio as key performance metrics. We use the average monthly T-bill yield during a year as the risk-free rate for that year in Sharpe ratio calculations. While we do not apply any stocks-cash switching frictions, we do calculate the number of switches for each scenario. Using the specified monthly data through October 2017, we find that: Keep Reading

Is market breadth a reliable indicator of future stock market returns? To investigate, we perform simple tests on four daily U.S. stock market breadth metrics:

1. RSP-SPY – Total return for Guggenheim S&P 500 Equal Weight (RSP) minus total return for SPDR S&P 500 (SPY).
2. NYSE A/D – Number of NYSE advancing stocks divided by number of NYSE declining stocks.
3. NYSE Up/Down Volume – Volume for NYSE advancing stocks divided by volume of NYSE declining stocks.
4. NYSE 52-Week Highs-Lows – Number of NYSE 52-week highs minus number of NYSE 52-week lows.

We use SPY as a proxy for the U.S. stock market. We use correlation tests that assume linear relationships between breadth metrics and future SPY returns and ranking tests that do not. Samples commence May 2003 (initial RSP availability) for the first three and late October 2005 for the fourth. Using daily dividend-adjusted levels of RSP and SPY and daily data for components of the other three breadth metrics from specified start dates through most of August 2017, we find that: Keep Reading

Does simple asset price trend following based on 10-month simple moving average (SMA10) reliably boost the performance of retirement portfolios? In their July 2017 paper entitled “Can Sustainable Withdrawal Rates Be Enhanced by Trend Following?”, Andrew Clare, James Seaton, Peter Smith and Steve Thomas compare effects of asset class diversification and trend following on safe withdrawal rates from UK retirement portfolios. They consider 60-40 UK stocks-bonds, 30-70 UK stocks-bonds and equally weighted UK stocks, global stocks, bonds, commodities and UK real estate (EW Multi-asset). They further consider risk parity (RP) multi-asset (each class weighted by the inverse of its prior-year volatility) and 100% global stocks (equally weighted across five regions). They focus on a 20-year retirement period (but also consider 30-year), assume annual withdrawals the first day of each year and ignore taxes and rebalancing frictions. They use both in-sequence historical asset returns and Monte Carlo simulations (random draws with replacement from the historical annual returns of each portfolio). They apply trend following separately to each asset by holding the asset (cash) when asset price is above (below) its SMA10. Their key portfolio performance metric is Perfect Withdrawal Rate (PWR), the constant real (inflation-adjusted) withdrawal rate as a percentage of initial portfolio value that exactly exhausts the portfolio at the end of the retirement period. Using monthly total returns in pounds sterling for the selected asset classes and values of the UK consumer price index during 1970 through 2015, they find that: Keep Reading

Does an asset class breadth rule work better than a class-by-class exclusion rule for momentum strategy crash protection? In their July 2017 paper entitled “Breadth Momentum and Vigilant Asset Allocation (VAA): Winning More by Losing Less”, Wouter Keller and Jan Keuning introduce VAA as a dual momentum asset class strategy aiming at returns above 10% with drawdowns less than -20% deep. They specify momentum as the average of annualized total returns over the past 1, 3, 6 and 12 months. This specification gives greater weight to short lookback intervals than a simple average of past returns over these intervals. Specifically, they:

1. Each month rank asset class proxies based on momentum.
2. Each month select a “cash” holding as the one of short-term U.S. Treasury, intermediate-term U.S. Treasury and investment grade corporate bond funds with the highest momentum. 
3. Set (via backtest) a breadth protection threshold (B). When the number of asset class proxies with negative momentum (b) is equal to or greater than B, the allocation to “cash” is 100%. When b is less than B, the base allocation to “cash” is b/B.
4. Set (via backtest) the number of top-performing asset class proxies to hold (T) in equal weights. When the base allocation to “cash” is less than 100% (so when b<B), allocate the balance to the top (1-b/B)T asset class proxies with highest momentum (irrespective of sign).
5. Mitigate portfolio rebalancing intensity (when B and T are different) by rounding fractions b/B to multiples of 1/T.

They construct four test universes from: a short sample of 17 (mostly simulated) exchange traded fund (ETF)-like global asset class proxies spanning December 1969 through December 2016; and, a long sample of 21 index-like U.S. asset classes spanning December 1925 through December 2016. After reserving the first year for initial momentum calculations, they segment each sample into halves for in-sample optimization of B and T and out-of-sample testing. For all cases, they apply 0.1% one-way trading frictions for portfolio changes. Their key portfolio performance metrics are compound annual growth rate (CAGR), maximum drawdown (MaxDD) and a composite of the two. Using monthly returns for the selected ETF-like and index-like assets over respective sample periods, they find that:

Keep Reading

A subscriber requested tests exploring whether a recent death cross for the Dow Jones Industrial Average (DJIA) portends an index crash. To investigate, we consider two ways of evaluating DJIA performance after death crosses and conversely defined golden crosses:

1. Behavior of the index during the 126 trading days (six months) after death and golden crosses.
2. Behavior of the index between converse crosses (death cross-to-golden cross, and golden cross-to-death cross).

We focus on distributions of average returns and maximum drawdowns (MaxDD) during specified periods. We also check robustness by repeating DJIA tests on the S&P 500 Index. Using daily DJIA closes since October 1928 and daily S&P 500 Index closes since January 1950, both through May 2017, we find that: Keep Reading

What makes trend-following tick? In the April 2017 version of his paper entitled “What Drives Trend-Following Profits?”, Adrian Zoicas-Ienciu investigates sources of trend-following profits in equity indexes and stocks. He focuses on daily trading signals for Dow Jones Industrial Average (DJIA) closing levels, as follows:

• Each day after the close, he compares the DJIA close to its simple moving average (SMA) plus or minus a buffer to suppress signal noise. If the close is above (below) the SMA plus (minus) the buffer, the signal is buy (sell). Otherwise the signal is neutral. He considers SMAs ranging from 2 to 250 trading days and signal buffers ranging from 1% to 5% for total of 1,245 rules.
• He implements signal changes at the next daily close by taking a 100% position in DJIA after a neutral signal, a (100%+x) position after a buy signal and a (100%-y) position after a sell signal. This approach allows separation of trend-following versus allocation effects. He assumes rebalancing friction 0.5% of traded value, cost of leverage (x) as the risk-free rate and return on cash (y) as the risk-free rate.
• He assesses rule performance principally as excess daily return versus buy-and-hold (B&H). He considers as alternative benchmarks the risk-free rate or a combination benchmark that is each day: B&H for a neutral signal; B&H for a buy signal; and, (100%-y) times B&H plus y times the risk-free rate for a sell signal.
• He assesses overall trend-following performance as the average performance of the 1,245 rules. He also considers the performance of an equally weighted portfolio of the top tenth (decile) of rules in each of 64 sequential 370-day subperiods.

He also evaluates the role of signal volatility (volume-weighted trading frequency) as a determinant of profitability. Using daily DJIA closing prices and 1-month U.S. Treasury bill (T-bill) yields as the risk-free rate during March 1926 through early October 2016, he finds that: Keep Reading