I have been using the $$HYIOAS symbol on Stockcharts for quite some time, but in a more primitive way. I use its trend as a simple measure of risk; when its 50 dma is above its 200 dma, then beware. Worked well for instance in '21, '22 and '23, but was noisy in '19 and unproductive in '14 and '15. Your QuantMage algo looks very promising, in contrast, as goes beyond a simple risk-on risk-off switch. I need to look into it!
As to Verdad's asset-class cycle diagram: if it also has a 17-year lookback, that would explain why it does not include an inflationary recession. Or the aftermath of an all-asset bubble where nothing but cash works. Both of which we had in 2022... I personally had a lot of IEF in '22 that I wish I had discarded in time.
Not at all to denigrate this excellent indicator, but beware of investing in treasuries or in hi-yield bonds if we get an inflationary recession crash in 2026.
A valid point! The strategy was in Overheating / Recession cycles most of 2022 (86.2%) and didn't perform very well for the year as you said (https://drive.google.com/file/d/1-2U8vVN6KbPT-7cUP0Th3h6Z5mLfUc0W/view?usp=sharing) since they don't have assets that can perform well during such an inflationary recession. I guess adding an asset like managed futures (e.g. KMLM) can be helpful for such periods.
'22 was a real outlier; to construct a strategy that will work in any situation is a tall order.
Even the classic Harry Browne portfolio w/one fourth each in cash, stocks, gold and treasuries didn't at all work well in 2022, and that was the first time since 1972 that this happened. To me, that is an indication less of the inflation problem, and more of the everything-bubble phenomenon.
Nonetheless, I'd seriously consider adding a simple trend filter to IEF. Only buy if it's in Golden Cross territory, for instance. With public debt so sky-high, treasuries are no longer the reliable and safe asset they used to be.
A trend filter is a good idea. I applied it to the Overheating / Recession parts, which slightly improved the overall Sharpe and turned its 2022 return positive. 2023 performance was somewhat traded off though. Check out the updated spell at https://quantmage.app/grimoire/4d568b1d176af34cc062c5f5c3a13f80.
Replicating Verdad's results is difficult because they use the ten-year trailing median value, which as far as I can tell, using Stockcharts, has since 1997 been anywhere between 6.7 and 4.38 (the latter being the current-day value).
In contrast, you use a fixed value of 4.5, which may be a good compromise, but perhaps loses out on a lot of effectiveness.
What difference does this make? Well actually, I think this can be quite substantial. In 2020, the trigger value was quite low, at around 5.0, so a risk-off signal was generated in a timely fashion (in February).
In 2009, in contrast, the trigger value was a lot higher, at around 6.2, so one re-purchased stock a lot quicker than one would otherwise had done.
In sum, a dynamic trigger enables the algorithm to respond to changing market situations a lot quicker, without getting noisier results or whipsaws.
Anyway, I continue to find all this quite exciting; I think I'll post a link on Bluesky to your article.
Thanks for sharing your experiment result! Yeah, using a fixed value can be disadvantageous. QuantMage unfortunately doesn't support a median, but it does support an average, so you can do something like this: https://drive.google.com/file/d/1obLMhpeA2DTV3yStCdbXaB2KZszmMENo/view?usp=drive_link As you can see, it shows a slightly worse, but still comparable performance.
This is great, thank you!
I have been using the $$HYIOAS symbol on Stockcharts for quite some time, but in a more primitive way. I use its trend as a simple measure of risk; when its 50 dma is above its 200 dma, then beware. Worked well for instance in '21, '22 and '23, but was noisy in '19 and unproductive in '14 and '15. Your QuantMage algo looks very promising, in contrast, as goes beyond a simple risk-on risk-off switch. I need to look into it!
As to Verdad's asset-class cycle diagram: if it also has a 17-year lookback, that would explain why it does not include an inflationary recession. Or the aftermath of an all-asset bubble where nothing but cash works. Both of which we had in 2022... I personally had a lot of IEF in '22 that I wish I had discarded in time.
Not at all to denigrate this excellent indicator, but beware of investing in treasuries or in hi-yield bonds if we get an inflationary recession crash in 2026.
A valid point! The strategy was in Overheating / Recession cycles most of 2022 (86.2%) and didn't perform very well for the year as you said (https://drive.google.com/file/d/1-2U8vVN6KbPT-7cUP0Th3h6Z5mLfUc0W/view?usp=sharing) since they don't have assets that can perform well during such an inflationary recession. I guess adding an asset like managed futures (e.g. KMLM) can be helpful for such periods.
'22 was a real outlier; to construct a strategy that will work in any situation is a tall order.
Even the classic Harry Browne portfolio w/one fourth each in cash, stocks, gold and treasuries didn't at all work well in 2022, and that was the first time since 1972 that this happened. To me, that is an indication less of the inflation problem, and more of the everything-bubble phenomenon.
Nonetheless, I'd seriously consider adding a simple trend filter to IEF. Only buy if it's in Golden Cross territory, for instance. With public debt so sky-high, treasuries are no longer the reliable and safe asset they used to be.
A trend filter is a good idea. I applied it to the Overheating / Recession parts, which slightly improved the overall Sharpe and turned its 2022 return positive. 2023 performance was somewhat traded off though. Check out the updated spell at https://quantmage.app/grimoire/4d568b1d176af34cc062c5f5c3a13f80.
Replicating Verdad's results is difficult because they use the ten-year trailing median value, which as far as I can tell, using Stockcharts, has since 1997 been anywhere between 6.7 and 4.38 (the latter being the current-day value).
In contrast, you use a fixed value of 4.5, which may be a good compromise, but perhaps loses out on a lot of effectiveness.
What difference does this make? Well actually, I think this can be quite substantial. In 2020, the trigger value was quite low, at around 5.0, so a risk-off signal was generated in a timely fashion (in February).
In 2009, in contrast, the trigger value was a lot higher, at around 6.2, so one re-purchased stock a lot quicker than one would otherwise had done.
In sum, a dynamic trigger enables the algorithm to respond to changing market situations a lot quicker, without getting noisier results or whipsaws.
Anyway, I continue to find all this quite exciting; I think I'll post a link on Bluesky to your article.
Thanks for sharing your experiment result! Yeah, using a fixed value can be disadvantageous. QuantMage unfortunately doesn't support a median, but it does support an average, so you can do something like this: https://drive.google.com/file/d/1obLMhpeA2DTV3yStCdbXaB2KZszmMENo/view?usp=drive_link As you can see, it shows a slightly worse, but still comparable performance.
Would trend clarity (or ppo?) help with identifying direction?
Trend Clarity indicates the strength of a trend, not its direction. PPO might be useful to catch direction changes, though.