Discretionary Models

updated Oct 10, 2008

The mechanical models did well... for a time. Their performance in '04 and early '05 wasn't much better than average, and made big losses in late '05 through '06. And, they're very volatile. Unexpected changes in fundamentals will always matter. And more, there are bright people in the business of "black box" design who constantly update their models. It's best to act as if being chased by a pack of wolves, in the words of hedge fund legend Jim Simons. I'm interested in seeing how long a newly tested black box can perform going forward. If the period of outsized performance is significantly longer than the time it takes to develop the model, then success should be sustainable. But for now, I'm focusing on models which have a large component of unquantifiable judgement and which incorporate news and changing fundamentals.

On Nov 11, 2003 I initiated the Diversified model; partly technical and partly fundamentals-driven. Before the Profunds trading time window closes, I look at all the basis indexes for the sector and broad market index funds of Profunds and use my proprietary technical signals, factor in news and other fundamentals, and then allocate among any of the sector and broad index funds for the following day. It's performance is outstanding - beating every one of the Profund leveraged funds. The chart shows the model has achieved this with much less volatility than any single Profund fund. The best Profund over this period has been the Oil & Gas Fund, shown in black. I will admit that it is extremely rare that I have used any of the short Profunds in this model - going short while the economic cycle is still in a boom phase (even if the stock market went nowhere in '04 and most of '05) is too risky. Still, I observe that when allocating for this model it allows me an easy excuse to get conservative, spread my bets into less favored funds, and often leave a chunk sitting in cash purely out of fear. It's resulted in a significant under performance relative to an "all in" bet, and so the next step in my evolution was...

On April 8, 2004 I went through the history of this model and made the Favored Fund model, which put the entire bet on the Profunds fund with the highest allocation within the Diversified model. After April 8, 2004, I gave myself the freedom to make the Favored Fund be any single Profund, not necessarily the highest allocation fund in the Diversified model. I expected the Favored Fund would outperform Diversified by a little, at the expense of much higher volatility. Instead, the volatility is still low, and the performance is much stronger than Diversified. It's further inspired me to think carefully about the mental state I put myself in when I make this decision. I've learned much from comparing that mental state to that when I make my Diversified decision, and trading my real money. And I note that now in the past couple of years, much excellent work has been done on the emerging field of "behavioral economics". Thankfully, there remains much inefficiency in how real people make real decisions, and how real people make real black boxes. This allows out performance by exploiting and fixing these inefficiencies. On this and following charts, I compare to the NASDQ 100 index, labelled NDX here. The performance is impressive.

These decisions are hard work and stressful; there are over two dozen funds (and growing) to consider, and the last 45 minutes of the trading day are the most important for measuring and decision-making. On Jan 25, 2005 I began a simpler discretionary model. All assets are placed in either the UltraOTC, the UltraShortOTC, or cash, depending on my best judgement considering technical, fundamental, and news-driven factors. Since the NASDQ 100 (NDX) is generally more volatile than the other broad index funds, this offers the greatest return to a strategy which can predict the moves more often than not. This is the Long/Short NDX model. It's doing remarkably well - giving Favored Fund good competition.

Model Applied to Wide Range of Profunds

At the same time that I make my decisions for the models above, I make a judgement on whether each of a selection of Profund sectors will pay off going long, short, or go to cash. Most sector Profunds do not have an associated 'short' version, so these judgements can't be actually implemented. Instead, it's an attempt by me to gather more data points in a given time span and get greater confidence that my judgement is actually good and not just lucky. From Sept 11, 2004 till mid '06 this is 899 trading days and over 10,000 decisions. These decisions have a batting average of 53% correct direction ('cash' decisions not counted) and the annualized return is 33.0% per year.

Performance - Since Inception to Oct 10, 2008