There were no position changes in the Country models this week. The Basic model continued to be dragged down by the selloff in treasuries but that may have found a pause or a base with the majority of the week’s price action being bounded by the first day’s range. The Stops & Targets model put in a slightly positive week but lagged behind its benchmark.
The SPY, after pausing around the highs of its recent trading range, went on to put in a strong new high close Friday off the heels of some positive news out of Europe.
This Week’s Strategy Lesson: ETF Model Drawdowns (Part 3)
We are now in the third part of this series. We opened up defining and discussing the types of drawdowns we see in the ETF models. Last week we examined in detail the probabilities of encountering these drawdowns in your initial equity. This week we are going to move beyond the historical track record and use a statistical sampling technique called the “Monte Carlo” method to see what we can learn about the inherent risk and return characteristics of these models.
Monte Carlo Simulation
The Monte Carlo method was invented in the late 1940s at the Los Alamos National Laboratory by Manhattan Project veteran and mathematician Stanislaw Ulam. While the method was initially used for physical and mathematical problems, it eventually made its way into modern finance.
The simulation works by first defining the rules or the calculations that will be performed. Then it randomly varies certain key variables according to their probability distribution. Finally, it runs the simulation using the random variables thousands (or potentially millions) of times and aggregates the findings leaving us with a potential range or outcomes and their probabilities.
For the purposes of our simulation here, the basic calculation will be tracking the models equity by taking the prior week total equity and multiplying it by a weekly return number 425 times (roughly eight years’ worth of data).
The weekly return is the “random” variable. To find this, we calculate the average weekly return (.43% for Complete) and standard deviation (2.1% for complete). The simulation uses these two points of data to randomly generate 425 different weekly returns that are consistent with the types of variables you would encounter that have an average and standard deviation as above.
It then uses these variables to generate (in our case) thousands of “alternative” histories for our models. What we see historically is what actually happened. For instance, the drawdown in mid-2012 was related to European debt concerns which were temporarily resolved (or pushed down the road) and the market recovered.
But what if a second negative event followed on the heels of the first. Or what if the leaders had been able to broker a deal before the situation reached critical levels. If anything different had happened, the ETF models could have been higher or lower or had different risk characteristics. By running thousands of simulations with different combinations of returns in different frequencies and in different orders, we can get some idea about the range of possible outcomes.
It should be noted that the distribution of possible values for our random variable (weekly returns) will be “normally distributed” in our simulations. While we know that markets are not strictly “normally distributed,” we can make progress towards overcoming this limitation by running a high number of simulations. This is, in part, the primary strength of this method of analysis.
ETF Complete Model
The results we show here are from running 10,000 simulations on the ETF Complete Portfolio data. The information is broken down into probability percentiles. The 462% 50th percentile result below basically says that of the 10,000 simulations, the median (middle) return was 462% with half of the simulations reporting a lower value and half reporting a higher value. It functions the same for the other percentiles groupings. For the 80th percentile, we are saying that roughly 20% (80th to 100th percentile) of the simulations returned a 713% total return or higher.
From this we can see that our actual return is fairly close to the median, 50th percentile return. We can also see that our historical maximum drawdown of 13% is smaller than the median 18% drawdown in the simulation.
The higher 50th percentile drawdown number means that we should probably use position sizing and trading guidelines that anticipate for the possibility of larger drawdowns than those that we historically experienced.
ETF Sectors Basic
Our weekly return variable for the Sector Basic model has about a 50% higher average return (.63% vs .43%) but over 100% higher standard deviation (4.8% vs 2.1%). This means that we should expect higher overall returns than the Complete, but that the range of possible returns will also be much wider.
Once again, our actual return and the 50th percentile return are fairly similar. It is generally good to see these numbers come out with similar values. That means we can’t say that we were particularly “lucky” or “unlucky.” The closer the numbers match, the more likely it is that our historical performance is repeatable and not some fluke of circumstance or rare occurrence.
Similar to the Complete, we also see that the 50th percentile maximum drawdown was a little higher than the one we actually experienced in the historical period.
ETF Country Basic
The Country Basic model has an average weekly return of .37% and standard deviation of 2.85%. The return is closer to the Complete but the higher standard deviation means the range of possible outcomes will be relatively wider.
The 50th percentile return of 300% is similar to our current return of 272%. Similarly, we also see that the actual drawdown was on the low side compared to what to what the simulations would predict as the most likely outcome.
Monte Carlo analysis provides us another way to think about the data and the risk and reward characteristics of our models. As with any mathematical/statistical method of looking at markets, it’s not perfect.
Markets involve numerous people and an almost uncountable number of factors from personal financial decisions to broader economic trends. As long as people remain unpredictable, markets will always have the ability to surprise us.
That is why it is always important to trade with the potential risks in mind. Combining both the historical data and the “what if” data from simulations gives us the most complete picture available and can help us make wise trading decisions.
The Current Condition of the Model
For the country model, we are in FXI, IFN, and TMF. FXI is currently in fourth place, and while we are up in this position, SSO continues to outpace it and we could see a position change soon if this trend continues.
Stay tuned to daily updates for any position changes.
Here is a summary of the weekly performance of all the ETFs that the strategy monitors.
Best wishes for your trading,
James Kimball
Trader & Analyst
MarketGauge
