My models for both SPY and BND are both on high alert, but what I'm trying to do is maximize after-tax returns, which means that I only sell when it is absolutely necessary. Given how much I had to take on the chin last week, I think it would be helpful to discuss what would have to happen to make selling "absolutely" necessary in my view.
The short answer is: "Not much." My bond model will go negative if, at the end of the month, the yield on 3-month bills remains at 0.5% and the 10-year yield closes above 2.68%, compared to 2.54% where it is now. Since the yield on 10-years rose to 2.54, from about 2.14 in just the past 5 days, the sell signal is just a hair breadth away. The stock model will turn negative if the S&P 500 is below 1560 at the end of this month, or about 2.9% lower than Friday's close.
Why Not Just Sell Now?
The stock and bond models I use include both value and technical components. The value components become negative as the market rises and positive as the market falls. The technical signals work in the opposite direction, becoming negative in down-trending markets and positive in up-trending markets. In both cases, the emphasis is on after-tax returns, so I try to minimize trading signals as much as possible.
Both stocks and bonds are over-valued according to the valuation components of the models, but we do not sell until the trend turns negative. There's a very good reason for this: The stock model began to indicate over-valuation at the end of July of 2009!. Stocks are up 73% from that point. This is not unusual and is why I started including a technical component.
The technical components use simple moving averages to determine the trend. If the actual price is above the moving average, the trend is considered to be positive. If the actual price is below the moving average, the trend is considered to be negative.
The sensitivity of the moving average is critical. If it is too sensitive, buy & sell signals will be susceptible to whipsawing actions, where the market turns just enough in one direction to trigger the signal, only to reverse immediately and cause a loss. If the moving average is not sensitive enough, the signal will come too late and miss out on major moves. Additionally, the more sensitive the signal is, the more trading it will generate per unit of profit, and thus, the higher taxes it will incur.
One of the innovations I've introduced relates to the multiple sensitivities that I use. The idea is that when an asset is over-valued, I want a signal that will be fast to get me out yet slow to get me back in. When an asset is cheap, I want a signal that is fast to get in, and slow to get back out.
For bonds, I use the total return of the 10-year treasury relative to the total return of the 3-month treasury as a base for generating the signal. This time series has moved in extremely long secular trends; the trends have typically last so long that since 1953, there have really only been on secular bear market and one secular bull market. This simply isn't enough data to create a statistically valid model, so we have to make educated guesses. There's no choice.
Given the extraordinary long length of the previous trends, however, I want to err on the side of inaction. On the other hand, we know that the nominal rate on the long-bond probably will not go to zero, so as yields approach zero, I feel comfortable increasing the sensitivity of the exit signal. The formula I've come up with worked in the last bear market in bonds, it was fairly early in calling the greatest turning point in the last century, and has kept me long throughout almost the entirety of one of the longest bull market runs in any asset class in history. The model I've chosen could very easily not be the optimal model for the next bear market, but the risk is not that we will sell too late, but rather that we will re-enter too early.
I use an 82-month moving average for the slow trend indicator and a 35-month moving average for the fast trend indicator. Even the fast indicator is pretty slow by the standard of most timing models. The optimal yield at which we switched to a fast indicator for sell signals is 4.5%. Any rate higher or lower than this did not perform as well in the previous bear market. Obviously, in the current market, we passed that hurdle rate a long time ago. In fact, the model switched to a fast signal on July 1, 2007, but even with the faster signal, there has not been a single sell signal during the subsequent six years.
This concept is illustrated in chart 1 below. The relative total return is shown in red. The slow signal is shown in black, and the fast signal is shown in green. The blue line in the lower part of the chart represents the value component. When this blue line is at 2 on the right hand scale, bonds are under-valued and the fast signal is used for entry, while the slow signal is used for exits. When the blue line is at 1, the market is over-valued, and the fast signal is used for exits, while the slow signal is used for entry. Otherwise, the slow signal prevails for both entry and exit.
The second chart shows the combined signal in black. Notice how rarely the red line crosses over this signal, yet, how precisely it does so at major turning points. As a result of last week's action, we are now as close to crossing over as we have been at all but a handful of brief instances since first going long in 1982. The sell signal will be triggered if, at the end of the month, the yield on 3-month bills remains at 0.5% and the 10-year yield closes above 2.68%, compared to 2.54% where it is now. Since the yield on the 10-year rose to 2.54, from about 2.14 in just the past 5 days, the sell signal is just a hair breadth away.
The stock model is more complicated and more robust. Valuation actually controls the weighting, while the technical overlay adjusts upward and downward from the core position. The valuation component upgraded to full weighting in November of 2008, and then back down to half weighting at the end of July in 2009 where it has remained since. The technical model's most recent change was when it went positive at the end of January 2012. The current tactical weighting then, adds 50% from each component to total 100% of the policy or benchmark weighting. The benchmark weighting is chosen on the basis of one's age and risk preference, which is explained in more detail here. For illustration purposes in the newsletter, I always use a 40-year old with normal risk preference as an example for determining the benchmark.
For the valuation component to trigger a sell, the S&P would have to rise 5.8%, to 1,700. This would lower our weighting by 50% but not take us entirely out of equities.
The technical component ratcheted up to a more sensitive signal at the end of January 2013 when the price to trend earnings ratio rose above 17x. If this signal were to trigger a sell, we would take our equity exposure down to zero because a negative technical signal always reduces the fundamental position by 50 percentage points. For this signal to trigger, the market would need to fall only another 2.9%, to about 1560.
This concept is illustrated in chart 3 below. The S&P price index is shown in red. The slow signal is shown in black, and the fast signal is shown in green. The blue line in the lower part of the chart represents the value component. When this blue line is at 2 on the right hand scale, bonds are under-valued and the fast signal is used for entry, while the slow signal is used for exits. When the blue line is at 1, the market is over-valued, and the fast signal is used for exits, while the slow signal is used for entry. Otherwise, the slow signal prevails for both entry and exit.
The fourth chart shows the combined signal in black. There are more whipsaws in this series than with the bond market, but I've done my best to minimize them.