The chart above is a screen grab of Jason Goefert's Buy/Sell Confidence Model (aka Smart/Dumb Money Confidence Model) 2001-2006. This is one of the few models that Goefert doesn't disclose his data sources, but my guess is he is using Rydex/Proshares Fund Asset Flows of index funds and/or CBOE Commitment of Traders data for stock index futures. I added the red and green bands in Photoshop to more easily visualize the model's behavior in relation to the S&P 500. The dark green bands indicate periods where the model was oversold, the dark red bands indicate periods of over-optimism. The lighter bands indicate periods one month after the model reversed course from an extreme level.
This chart is highly instructive of how most sentiment indicators behave.
- Market indexes can continue to rise/fall after models reach extreme levels
- Model reversals are a more reliable indicator of a market reversal
- Sentiment models do not always reach extreme levels at market tops/bottoms. They aren't closed buy/sell timing models in and of themselves.
I use sentiment models/indicators to counter simple trend indicators. If you've ever experimented with technical indicators such as moving averages you'll know trend indicators are always wrong at both peaks and troughs. By balancing trend indicators with sentiment indicators you can create a timing model that is "less wrong" at key turning points. For example, instead of being 100% in cash at a market bottom, your combo model might suggest you only be 70% to 80% in cash.
I'm not looking for perfection here. You can see that although this model is a great tool, it isn't perfect...and it certainly cannot predict market prices. The purpose of a timing model imho is to reduce portfolio volatility, not to deliver outsized gains. If it ends up doing the later, great...consider it icing on the cake.
6 comments:
tom, keep teaching....great stuff...do you look at specific sentiment and price charts and then use points?
http://tinyurl.com/yota32
This is the sentiment indicator that I have been using. Trix top is a buy on the dip and trix cross zero upward is a sell. One is faster to look for divergences. Scale in or scale out is how I react. On the same page with you about goal of timing.
jasper
Yes, my timing model is composed of 4 trend indicators and 4 sentiment models. Each gets a score based on a set of defined rules. I run through the entire thing in due time.
Trix and P/C ratios are good sentiment indicators - Goefert uses variations of each. I prefer to use models composed of more than one indicator though.
Terrific Tom...I agree, better to have multiple points of data.
Tom, re my quest to backtest your model. Could we sample multiple weeks across a long time span...get all the highs, lows, and misc in between...and then see the hit rate? Not sure how easy it is to calculate points per your model with historical data. For me this would be a huge accomplishment to know to what degree to use my own mechanical signals on etf relative strength selection. Any steps to take the stress of emotions out of it is a good thing for me. I'm tired of being distracted by dissecting the dark side of the street. Fascinating but not good for port performance. jasper
jasper,
the problem with backtesting this model is with the sentiment side. You could probably test a composite of my trend indicators but it's only half the model. I switched from using my own sentiment models to sentimentrader in late 2005.
I keep timing separate from equity selection. My method is essentially:
1. Determine level of equity exposure
2. Of that equity exposure, determine the allocation between U.S., EAFE, and Emerging.
3. Lastly, determine the exact sectors, regions and countries in which to invest.
tom, I was thinking of using whatever composite model you have now and compare that to single benchmark index, assuming that the rise and fall of domestic and foreign mkts have some correlation, even if not in severity of rise and fall. I'll be content to patiently see the unfolding of your model and be open to understanding that I may be making some false assumptions about how to backtest.
Post a Comment