Sunday, December 9, 2007

Market Sentiment System Revisited - Part I

Last month I posted a two-part article building on the always excellent work of Dr. Brett Steenbarger of TraderFeed fame. The article presented an outline for an intermediate-term Market Sentiment-based swing trading system incorporating the observations of Dr. Steenbarger. As background, links to the article follow:

Part 1Market Sentiment Oscillator

Part 2Sentiment Trading System

With the closing of the open trade referenced in Part 2, I thought I would post the results of that hypothetical trade and revisit the original system to see how it might be improved.

First, November's trade opened on the 5th when the ten-day oscillator first fell below 0.995, and closed on the 30th when the 2-day RSI exceeded 90 for a total of 18 open trading days. This was about seven days longer than the average trade, as previously tested back through January 2003 (about 1,240 trading days).

Using the S&P500 (SPY) as the trading vehicle, the maximum drawdown of this trade would have been -6.1%, the worst of the entire test period. This comes as no surprise, as the test period represented a remarkable bull recovery right up until the recent stall. Nevertheless, due to the strength of the end-of-month bounce, by exit day the trade was down by only -0.7%.

This is the general nature of the trade: buying on extreme weakness in measured sentiment, and selling into price strength. However, as I wrote, "this simple system can generate long holds during very rocky times...." This trade was certainly no exception!

Second, the lesson to resolve was clearly how to avoid getting in too early. In this regard, I decided to take a closer day-by-day look at the ten-day oscillator in conjunction with the absolute value of its underlying indicator, as follows (the x-axes indicate days prior to exit):

As shown above, like many technical indicators, the best time to take the trade signal is when a divergence between the oscillator and its underlying value is observed, suggesting a change in momentum. Had the hypothetical trader waited for such a divergence to occur on or around the mid-point of the trade window, as indicated by the originally specified system, a significantly better result of +2.1% to +4.9% may have been achievable (see end-of-day bars -10 and -7, respectively).

Visually reviewing prior open trade periods, I made similar divergence improvement observations. This inspired the following rule modifications:

Buy – Buy the SPY at the close when (1) the oscillator has fallen below 0.996 during one the prior three days, and (2) the two-day RSI of the SPY is below 50. These rules are somewhat looser than before, but two additional criteria are added to pinpoint a potential divergence point, including: (3) the oscillator must be higher than its lowest close within the last three days; and, (4) the underlying indicator must be lower than its highest close within the last three days. In other words, the oscillator must have potentially bottomed, and the underlying indicator must have potentially topped.

Sell – Sell after the earliest of the following: the oscillator reaches 1.01, the 2-day RSI of the SPY exceeds 90, or twenty days has passed. This is the same as before.

During the test period, this modified system would have generated a 53% gain before trading costs with 28 wins out of 31 signals (90% wins/ average +1.7% gain), and was invested 21% of the time for an average hold period of eight days with a maximum drawdown of -4.3%.

These results demonstrate the desirable attributes of a lower drawdown, a higher win rate, a shorter average trade duration, and less total market exposure as compared to the system's first iteration. Indeed, the most recent November trade would have recorded a +2.3% gain. Even better absolute results may be garnered by waiving the higher oscillator criteria (rule #3) and just waiting for a confirmed capitulation of the underlying absolute indicator (total gains of 65%). These results challenge the total gains of the SPY over the same period (+79%) with substantially less market exposure.

This analysis is still "Level 2" at best, and more work is required before I would consider the system tradable per se, including a parameter sensitivity test. In a later article, I will also take a look at how the system might be used, if at all, intraday.


No comments: