I have noticed when I run a longer term (15+ years) backtest on a strategy that some ETFs exhibit a large drawdown during the first year or so after the ETF's launch. See attached screen capture.

Two that I recently come across have been EWW and XLE. I am trying to figure out why this might be. A few things that have some to mind are:

1. During the first few years the ETF can be very illiquid compared to years after launching.
2. The fixed slippage that I use per shares is a much large % of the share price early in the ETF's life when the price was very low.
3. The market was just fundamentally different in the period in question (1999-2002) than is was subsequent to that period and the strategy just plain doesn't work in that period.

Any more ideas?

VanderTrader 

I have noticed when I run a longer term (15+ years) backtest on a strategy that some ETFs exhibit a large drawdown during the first year or so after the ETF's launch. See attached screen capture. Two that I recently come across have been EWW and XLE. I am trying to figure out why this might be. A few things that have some to mind are: 1. During the first few years the ETF can be very illiquid compared to years after launching. 2. The fixed slippage that I use per shares is a much large % of the share price early in the ETF's life when the price was very low. 3. The market was just fundamentally different in the period in question (1999-2002) than is was subsequent to that period and the strategy just plain doesn't work in that period. Any more ideas?

Are you first optimizing on the latest data, and then applying the optimized parameters to all the data?

In the example I posted I optimized on all of the data.

VanderTrader 

In the example I posted I optimized on all of the data.

I would check that your slippage estimates were accurate. Could it be that back in 2003-4, the bid ask spread was tighter than you assumed in your test?

I would also look at volatility in that timeframe - was it much different?

If you are using position sizing in your strategy, that could be biasing results too.

Finally, is your system biased to long or short side at all? That might influence your results.


Without seeing exactly how you are testing, and what exactly your strategy is, it is hard to say, what, if anything, is really going on. You may be doing something wrong, and you may not be. I've certainly had this occur before. Sometimes a system just doesn't work well during a certain market period.

All good questions and suggestions, thank you, Kevin. I use the same strategy using other ETFs which were very liquid at the time (IE: SPY) and the same strategy did very well in those years.

So, I keep coming back to the liquidity, or lack thereof, of these ETF in the first few years after launch. Here are two screen shots showing charts with 83 minute bars. You can see how illiquid they were during those years.

Since the data is so sparse, and some days have more bars than others, I would not trust results at all. Any calcs you do will be totally messed up by missing data.

Thanks again Kevin.

PS: I am looking forward to the launch of your book!