I have read the following article and tried to replicate but with a half baked success:
Here are the key points:
- Custom Universe with 300 largest stocks in Canada;
- 20 stocks rebalanced on a yearly/monthly basis;
- Stocks must pay a dividend (Yield>0);
- Pick 20 stocks with the lowest annual volatility;
I calculated the return from 1999-2019 to make a fair comparison because that’s the most historical data we have access. His CAGR is 12.66% for that period. I’m getting something similar (see portfolio simulation below). However, when I turn this into monthly rebalancing the performance turns bad. Does anyone know what’s missing to replicate the article’s results?
https://www.portfolio123.com/port_summary.jsp?portid=1574339
Thanks!
There are two reasons for the discrepancy.
-
When you rebalance yearly, your turnover is 0.75X. When you rebalance monthly, it’s 2.63X. You’re often paying 50 basis points in transaction fees, because these are not very high-volume stocks and you’re using variable slippage, and occasionally it’s even more; for example, for CNU you’re paying 500 basis points. So with round-trip costs of about 1.3% per stock, you’re paying altogether about 3.5% per year in transaction costs.
-
The starting point is EVERYTHING in a 20-stock model with a yearly rebalance. If you start in January 1999 you get a nice 13% CAGR. If you start in July 1999, you only get a 9% CAGR, using exactly the same system.
Use this version instead.
https://www.portfolio123.com/port_summary.jsp?portid=1574382
Note that in the universe you had it set to All Fund Canada. I set it to TSX as per the article. This probably hurt performance however and didn’t improve it.
You were using daily beta as a proxy for volatility. I have not found beta to be a very good replacement for volatility. I use it to enhance but not replace the signal. My preferred factor is ATRN or Average True Range Normalized. It only has 100 look back days but that is enough.
I removed transaction fees as I don’t believe he used any. The goal is to replicate his system as close as possible and you can add fees back in later once you feel it jives.
I raised the threshold of dividends to a yield of 3%. You can alter it but basically everything in Canada has dividends so I wouldn’t place much emphasis on a system called Canadian Stable Dividends if the the dividend requirement is only to have one. Basically, it is a volatility system which does work well in Canada. But I suggest raising the threshold of dividend yield if it is in fact a Canadian dividend system.
I added in a sector weighting rule.
Hi Hemmerling,
Thanks alot for your input, I appreciate it. Here are my comments on your improvement:
-
Very good thinking for using the ATRN. However it is not limited to 100 days look back anymore. Try 252 days, it doesn’t return an error. I believe Marco increased the limit a couple months ago when people were complaining about the arbitrary limits on technical indicators.
-
Yes you are right he didn’t use any transaction fees that’s for sure. However that’s a major flaw, I wouldn’t present a model portfolio strategy in the newspaper with that kind of omission. By removing them, it gets closed to what he has done for sure.
-
Your point is totally valid. It is more of a volatility ranking system than it is a dividend one unless you state a minimum level, just like you did.