A bit calmer, up 0.86% for the week. Interesting that bonds have increased, mostly coming at the expense of gold and cash. The “defensive assets” are still providing a headwind though. On February 5th, the strategy rebalanced to:
38% SPY , 38% TLT , 19% GLD , 5% Cash
I’m hopefully two weeks away from automating this update fully on its own page of the blog. I need to test it a bit live first before switching it on.
*All investing strategies come with the risk of loss, including this one. This portfolio may not be appropriate for your investment goals and requirements, and it is not investment advice.
It should not be assumed that recommendations made in the future will be profitable or will equal the performance of the securities in this list.
Calculations are my own. Returns shown do not include trading costs. They do not include any fees. Past performance is not indicative of future performance. Dividends are re-invested.
How much data do you use to make your calculations? I’ve been playing with the numbers and so far I would have had a better performance using 90-180 days back. Saludos.
I’m biases to shorter timeframes. https://breakingthemarket.com/convergence-time/
There are different timeframes that work better historically than what I’m using. I don’t backtest lookback times, but try and mathematically and conceptionally justify the choice. Its too easy to say “73 days” is the optimal lookback period from a backtest, when there isn’t any reason why 73 is the best going forward.
Hi BTM!
I absolutely love the blog and have a few questions if you don’t mind.
1. Do you use implied volatility in your standard deviation calculations? Intuitively it seems like it should be better at predicting near-future volatility compared to historic volatility. Or perhaps averaging them is best. Backtesting with IV may be impossible though as I’ve not seen any data on it going beyond the last decade.
2. How do you decide the return on gold? The only correlations I could find are extremely minor ones with low interest rates and with extremely high inflation.
I think a pragmatic approach could be to predict minor returns when 1Y treasuries are below the 5-year average and additional returns if inflation fear (https://fred.stlouisfed.org/series/T10YIE) is above the 5-year average. I would never feel comfortable predicting returns over ~3% though.
3. Do you have any thought on Damodaran’s implied ERP calculations (http://pages.stern.nyu.edu/~adamodar/). I found them to be shockingly accurate when I backtested them against future 10Y returns of the S&P. I’m not sure how accurate they’re going to continue to be in the future but I assume they’re more accurate than any calculations I could come up with on my own.
4. I’ve seen you write a few times that TLT is linked to the 30Y treasury but isn’t it more-so linked to the 20Y? While TLT’s average maturity is 26.1 years, shouldn’t the average maturity (of 18.8 years) be more important for pricing the fund as its what influences valuation and price movements?
Thanks!
1.) No, but I don’t hate the idea if you want to do that. IV has an inherent bias in it to be slightly higher than future realized vol, which isn’t necessarily a bad thing as it will keep you conservative and help you “miss left”, but it isn’t really a prediction of the future vol. There is also the back testing problem that you pointed out.
2.) Gold is very hard to predict. One thought is it should be similar to the growth of the money supply. I’ll talk about it in a later post. I like meathods similar to the one you describe, but haven’t ever felt fully comfortable with them.
3.) Its an interesting calculation, but it seems a bit recursive. At some level he’s just saying it will be similar to what it was before. He’s also calculating the geometric premium, not the arithmetic, which is fine, but is of course smaller. But I should study it more.
4.) TLT may be closer to the 20Y than 30Y. I’ve always though it was roughly in between and I often get the difference between maturity and duration confused. Utltimatly this is a difference in backtest vs real life though as I don’t size my TLT exposure based on the 30Y data, but on what TLTs own vol and corr.