We frequently review and revise our trading models to find ways to improve results based on sound reasoning and data-driven justifications. However, it is important to not add too many parameters with precise settings to avoid curve-fitting of historic datasets. One way to accomplish this is to compare model performance with out-of-sample data to expected in-sample results. Sometimes this is as simple as optimizing the model parameters for a time period (e.g. 2015-2020), applying the same parameters to a prior time period (e.g. 2009- 2014) and then comparing the results for consistency.
This week we retrieved analysis performed for a presentation made to a medium-sized hedge fund in Chicago in September 2014 to see if recently observed patterns can be validated by prior time periods. In this case, the sum of the Dow Jones 30 Index gains on Fridays so far this year are negative 2.3% versus a positive 5.5% for the other four days of the week. The absolute ratio of these two numbers is 2.4 compared to the same ratio applied to the dataset below for a period of June 2000 through June 2014 of 2.9. So, there are two independent datasets suggesting that the gains in the DJ30 during the first four days of the week are around 2.5 times in size as the losses that occur on Friday – on average and in general.
We rigorously tested this theory applied to our daily TQQQ 3X leveraged Nasdaq model and did find that it improved results by around 10% if we exited the position on Thursday at market close and then re-entered on Friday market close if the position would have otherwise been open at that time or provides a buy signal for the open on Monday. However, we were not able to validate this theory for the UPRO 3X leveraged S&P500 ETF model, so we’ll leave that logic out which will provide some measure of diversification by time as one model will be active on Friday and apply new buy signals on Monday at open, and the other will be dormant on Friday and apply new buy signals on Friday at close.
It’s worth noting that the above excerpt from the 2014 PowerPoint presentation indicates that a simple strategy of holding the Dow 30 tracking ETF (DIA) from Friday close to Thursday close and then swapping this for the inverse ETF (DOG) on Thursday close to Friday close may provide results 2X the buy-and-hold DJ 30 strategy (more fodder for your next “you can’t beat the market with active trading methods” conversation).
Lastly, we concluded a multi-month study comparing models based on 15-min data updates versus models using daily data and found that there is not enough (if any) added benefit to using the more complex and time-consuming 15-min models, so these are retired for now.
Forte Strategy Update
We executed 4 trades last week for a net gain of 0.6% compared to a loss of 0.6% by the S&P 500. Our YTD net results equal a 3.6% gain compared to a 2.7% YTD gain for the S&P 500. Our YTD max drawdown is 9.5% versus 33.9% for the general market.
More details about our trading activity can be found by registering on the Collective2 website and searching for Forte Strategy. A running list of these email blogs and general information about Maestro Capital Research can be found at maestrocapitalresearch.com.