Notice: Beginning October 16th, 2007 this system which is described below will begin using full margin, 4% of QQQQ price stops and may include cash positions at times. Everything else below in the description still applies.
TMG is an Excel-based quantitative analysis trading program that utilizes a proprietary rule weighted algorithm consisting of numerous price-based and breadth-based technical indicators to generate a daily signal for trading the Powershares QQQ ETF (Symbol: QQQQ, formerly Nasdaq 100 Trust) and related derivatives. The system backtesting maximizes the historical and statistical correlation to Truth for optimal performance for a once daily EOD signal/NDO trade against the NDX since Jan 2000. Signal length varies from one to several days, and over the backtest period, there have been 854 round trip trades, so commission costs, while not reflected in the above numbers, should not be excessive. Algorithm enhancements are made periodically as a part of an ongoing continuous process improvement program. System returns are characterized by normal statistical variations over the backtest period and therefore going forward are non-deterministic. Subscribing to TMG and trading the signal is not a short term, get-rich-quick strategy, and it is only over time that the historical statistical results should be realized. When the next day's market direction is affected by emergent economic announcements and world geopolitical events, the odds of a False signal are greatly increased, since those events frequently cause the market to react in a way that is counter its normal statistical indicators. What this means to a potential subscriber is that over any arbitrary time frame that you might choose (e.g. the period of your membership), returns may be greater or less than the statistical norms. To avoid unrealistic or unrealized expectations, every prospective subscriber should carefully consider whether or not they possess the patience and discipline to stick with the system through flat-to-down periods that the historical statistics virtually guarantee will occur. To illustrate, if a fair coin is flipped 100 times, how many times could one expect to get four heads or four tails in a row? The first flip in a sequence can be a head or a tail, but the remaining 3 flips would need to match the first flip. Therefore the probability of getting four heads or four tails in a row would be 1x(.5^3)=.125, and the streak could happen anywhere from the first four flips to the last four (1,2,3,4 up to 97,98,99,100), which would be 97 opportunities. Accordingly, the expected value would be the probability of a single occurrence multiplied by the number of opportunities, or ~ 12.3 such occurrences (i.e. 4 heads or tails in a row) in 100 flips. If you limit the outcome to only heads or only tails, then the expected value drops to ~ 6, i.e. (.5^4)x97=6.1. Based on the backtesting, the single day win probability is ~ .7389, and therefore the loss probability is ~ .2611, which says that you could expect a losing streak of 4 or more ~ every 174 days of trading, and depending on the market volatility, 4 or 5 incorrect signals in a row can result in drawdowns in the equity curve that may be significantly more than any single trade stop loss, since the losses would be cumulative.
The equity curve shown reflects the benefit of several improvements that have been made in the algorithm over the past year since going live. In stating the system performance, we believe it is more important to focus on the backtested returns of the current version, since we feel that those should be more representative of performance going forward. The downturns we have experienced (and learned from) have been primarily the result of periods of extreme short term volatility characterized by successive daily or more frequent market reversals that have followed major disturbances in the market, and for which a once-a-day signal/trade system is not well-matched. Even though similar periods have occurred previously in the backtest and the backtesting showed that the system eventually recovered, there appears to be little or no tolerance by many subscribers for trading through them in real time with real money, so if you are not prepared to deal with those, you should save yourself the trouble of subscribing. In almost eight years of backtesting, there have been 2 instances of 5 incorrect signals in a row, while there have been 115 instances of 5 or more correct signals in a row, including 1 correct string of 23. By logical extension of the Nyquist/Shannon Sampling Theorem, one must sample (and be willing to trade if necessary) at twice the highest frequency of the market reversals in order to capture them accurately, which of course is simply not possible with a once-a-day signal/trade approach. Going forward, when these periods of high volatility are detected, the signal will still be published, but a recommendation will be made to go to Cash until the extreme short term volatility has moderated.
In spite of the charges and accusations of a few disgruntled subscribers who apparently do not understand the statistical realities of trading a quantitative model and only demand that the system performance match the backtested returns over any arbitrary time frame of their choosing, the numbers we have published relative to backtested returns are absolutely true and accurate. Moreover, we are willing to make a backtested trades file available to anyone who requests it in advance of their signing up so that they may verify the numbers for themselves.
News 
Market Data 
Discover 
