After writing the post Timing The Market With Perceived Credit Risk, a colleague pointed out some of the short-comings of this market-timer, and also that a simple timer using the unemployment rate produces better results.
I haven't had a great deal of time to investigate further but have decided to lay out my thoughts on this matter. First of all, I never really considered that the timer shouls capture every market turning point perfectly. I was (am) happy that it identified most of the market troubles, including the crash of 1987, the Asian financial crisis of 1997-1998, the bear market of 2007-2009, and the corrections of 2011, 2015, and 2016.
The timer missed much of the dot com crash. Part of the reason for the missed signal was due to the fact that the TED Spread was already quite high leading up to 2000, possibly due to the 1998 financial crisis, collapse of a large hedge fund, and Y2K concerns. The algorithm that I specified is such that the TED Spread needs to be rising in order to declare "Risk Off", not falling. The absolute value of the spread was still above 0.5 for the first part of the dot com crash, but not all of it. I will likely include a threshold of 0.5 as part of the timer logic in addition to the EMA(50)>EMA(200), a slight improvement.
Other than missing one major bear signal, the timer switches to "Risk Off" for some periods of time during major market expansions during the 1990s and 2003-2006. This is partially due to the nature of the algorithm... I would expect the "Risk On" to "Risk Off" ratio to be 1:1. For me, it is not a major problem as the algorithm is identifying periods of risk, not downward market movement. If I ultimately decide to create a long/short indicator then the TED Spread will be part of it, but there will be several other signals used in tandem.
There is an abundance of possible timing indicators that could be used for market timing. I don't want to list any significant details yet but below are two in the early stages of development.