Author Topic: Gold price and the hazards of backtesting  (Read 8960 times)

Radagast

  • Handlebar Stache
  • *****
  • Posts: 1577
  • Location: West of the Mountains, East of the Sea
  • One does not simply work into Mordor
Gold price and the hazards of backtesting
« on: August 23, 2017, 12:48:39 AM »
Many people who have played around with back testing portfolios have come across gold as an option, most prominently at https://portfoliocharts.com, but also at https://www.portfoliovisualizer.com and https://cfiresim.com. If you play around enough, the results from the 1970's may start to look suspiciously good in a time when stocks and bonds both crashed. Some people may then go on to read about the US leaving the gold standard or something about "Bretton Woods "and assume that was the primary cause. Other people may assume that gold, like other real assets, did well in a time of high inflation. I was curious, so I did some investigation.

Below is a table of the price of gold since 1954, and also the consumer price index, both taken for December of each year. There is also a December 2016 inflation adjusted gold price. Finally, the table has the price of gold divided by the consumer price index. This is a handy little trick I got from an article by Larry Swedroe http://www.etf.com/sections/index-investor-corner/swedroe-dont-be-distracted-golds-glitter . It is a convenient metric to adjust for inflation because it is independent of the year. I also looked up the cost to mine gold, for example at Barrick http://www.barrick.com/operations/pueblo-viejo/default.aspx. Barrick's all-in cost of production at their lowest cost mine might be regarded as a "book price" for gold, below which the price is unsustainably low (though with the waste from copper and silver production, and the huge amount already mined, it could presumably stay low for a decade or more before inevitably rising again to a cost which could support dedicated gold mining).


(Note: Barrick's lowest US mine and lowest overall mine numbers are switched, the US should be the higher number. I didn't feel like fixing it after I noticed)

You can clearly see in the table the years when the US stopped fixing the price of gold. The price was low and steady through 1972. In fact, the US government fixed the price of gold at such a low level that the market has never assigned it a similar value since it was free to do so. Furthermore, the price of gold in 1970 was essentially the lowest real price in all of history, if you believe this: http://www.zerohedge.com/news/charting-price-gold-all-way-back-1265 (zerohedge, I know). It was also apparently well below the cost to mine more. Since gold prices were allowed to be set by the market beginning in 1972 (I am actually having a hard time figuring out exactly when this happened, was it actually in 1973?), they have been much higher and far more volatile.

What's the big deal? Well, many common back testing tools used by Mustachians begin in 1970-1972. Even more coincidentally, that period was a stock market peak before the 1973-74 bear market, which was one of the worst in US history. Now lets see another graph (different data set) which shows the price of gold and the price of the S&P500 relative to the consumer price index.


Aren't markets efficient? After the US officially left the gold standard, it took the market only a few months to nail the average real price of gold over 1973-2016. By March 1974 gold had attained its subsequent real average value. And oh-so-coincidentally, the the trip from government price to market price occurred simultaneously with the stock market crash to such an extent it is almost creepy.


Finally, here is another graph demonstrating the performance of gold relative to Wilshire total stock market total returns (dividends included). Stocks look a lot less scary here. I also threw in Barrick's cost to mine. If you take the mining cost as the actual price of gold, then it is fair to say gold had reached its "correct" price by June 1973.

Conclusions:
  • 100% of the price movement of gold (and thus hypothetical investment return) from before 1972 to March 1974 can be explained by the government fixing the price and the market subsequently finding its real price. No "flight to safety" phenomena is necessary to explain its return over this interval. Based strictly on the history of gold prices, there is no reason to believe this sequence resulted from an external economic situation and is thus repeatable.
  • This period of price determination coincided very precisely with the stock market crash of 1973-74.
  • Back testing should not be performed using gold as an asset class until March, 1974 when it attained its subsequent real average market value. Some back testers may find it acceptable to push this back to as early as June 1973 when it reached a cost which could sustain mining (at 2016 mining costs).
  • Back tests of gold beginning in early 1973, 1972, 1971, 1970, or earlier should not be regarded as valid. This is especially a critique of Portfoliocharts which presents many years of data into a single display. That in itself is not a problem, however, Portfoliocharts tends to emphasize the worst case returns, all of which would have been greatly mitigated by owning gold before the market could set its price, and rebalancing after it had done so. Many Portfoliocharts calculators also tend to obscure exactly which lines started in invalid years. Furthermore, it hatches out questionable years for many assets, but not for gold.
  • I also provided a bunch of other information, which may be useful to potential gold investors.
  • Back testing gold before March 1974 is bad, and wrong. There should be a stronger word for it. Like badwrong. Or badong, for short. We should stand for its opposite: gnodab.

steveo

  • Handlebar Stache
  • *****
  • Posts: 1944
Re: Gold price and the hazards of backtesting
« Reply #1 on: August 23, 2017, 03:51:12 AM »
Great post. I've said previously on this forum that people that use these asset allocation tools are probably doing themselves a disservice. The trick is to realise that the past had unique opportunities that you probably couldn't even obtain and that probably won't occur in the future. Therefore don't try and come up with the perfect asset allocation based on historical data.

I honestly think the smartest asset allocation is own your own home and invest in international stocks and domestic bonds as per your risk profile. It's the smartest because it's the simplest as well. I don't though follow my own advice.
« Last Edit: August 23, 2017, 03:53:30 AM by steveo »

SeattleCPA

  • Handlebar Stache
  • *****
  • Posts: 1442
  • Age: 59
  • Location: Redmond, WA
    • Evergreen Small Business
Re: Gold price and the hazards of backtesting
« Reply #2 on: August 23, 2017, 05:51:52 AM »
I agree. Great post. Good work, Radagast.

I'm going to disagree with Steveo though... I think those backtesting tools are good and that trying to take a Modern Portfolio Theory approach to portfolio construction makes sense. (I.e., riskless assets, trying to include asset classes with lower correlations, etc.).

The problem I see is there isn't really data available at portfoliocharts.com, portfoliovisualizer.com and cfiresim.com to look at the historically worst case scenarios you want to consider. The first two tools mentioned have lots of asset classes (kind of) but only start in 1972 and so leave off those worst case scenarios you and I want to consider including the one, "you retire starting in 1966." cfiresim data starts in 1872 but you've only got three asset classes, well, and then cash.


GenXbiker

  • Bristles
  • ***
  • Posts: 327
Re: Gold price and the hazards of backtesting
« Reply #3 on: August 23, 2017, 06:16:22 AM »
I've seen this info mentioned a few different times recently regarding those back-tested portfolios containing a gold component.

Tyler

  • Handlebar Stache
  • *****
  • Posts: 1158
Re: Gold price and the hazards of backtesting
« Reply #4 on: August 23, 2017, 07:59:46 AM »
Back tests of gold beginning in early 1973, 1972, 1971, 1970, or earlier should not be regarded as valid. This is especially a critique of Portfoliocharts which presents many years of data into a single display. That in itself is not a problem, however, Portfoliocharts tends to emphasize the worst case returns, all of which would have been greatly mitigated by owning gold before the market could set its price, and rebalancing after it had done so. Many Portfoliocharts calculators also tend to obscure exactly which lines started in invalid years. Furthermore, it hatches out questionable years for many assets, but not for gold.

Thanks for your detailed and thoughtful gold explanation! 

I'll simply point out that the majority of the Portfolio Charts visuals are specifically designed to allow the user to see the big picture while discerning certain start years for any reason you like.  For the Heat Map, just ignore individual rows.  And for any chart with multiple lines, note that they are color coded by start date with the oldest lines darker than the newest ones.  The "hatching out" of data is solely related to assets where I've replaced the desired one that has no good data with an alternative similar asset, while in this case the data is fine and the issue is a matter of interpreting the numbers.  But now that you mention it, I can perhaps look into how to expand that functionality to isolate individual years for situations like this.  My goal is to provide as much good information as possible, not to obscure anything. 

Feel free to PM me if you have any suggestions.  I'm always open to new ideas.
« Last Edit: August 23, 2017, 10:42:25 AM by Tyler »

SeattleCPA

  • Handlebar Stache
  • *****
  • Posts: 1442
  • Age: 59
  • Location: Redmond, WA
    • Evergreen Small Business
Re: Gold price and the hazards of backtesting
« Reply #5 on: August 23, 2017, 09:21:41 AM »
Tyler, I love your work at www.portfoliocharts.com. Absolutely love it. Great contribution to the larger discussion.

Further, I particularly appreciate the work you've done with withdrawal rates for various asset allocation formulas. (I want to do a blog post about this at some point...)

But the one thing that makes me a little bit anxious (and this is absolutely NOT a criticism of your site) is that the withdrawal rate calculations don't include the nightmare scenarios like retiring in 1966.

Another way to say this same thing: I'd love to work with the SWR your site shows for a Swensen portfolio... but gosh I wish there was about 100 more years of data!

Tyler

  • Handlebar Stache
  • *****
  • Posts: 1158
Re: Gold price and the hazards of backtesting
« Reply #6 on: August 23, 2017, 10:51:57 AM »
Another way to say this same thing: I'd love to work with the SWR your site shows for a Swensen portfolio... but gosh I wish there was about 100 more years of data!

So do I!  :)  I'm always on the lookout for more data history and will be happy to add it as I find it.  BTW, you might check out the Withdrawal Rates FAQ if you haven't already. 

But enough about Portfolio Charts -- Let's get back to Radagast's excellent gold discussion.

I'll certainly concede that looking at data alone often doesn't tell the whole story.  However, I don't think the problem is isolated to gold by any means.  For example, Robert Shiller argues in "Irrational Exuberance" that the US stock bubble in the 90's was just as much of a historical aberration as gold performance in the early 70's.  No backtesting tool will ever be a substitute for good sense, and I appreciate discussions like this that help put some of the numbers into proper historical context.  Just be careful not to add too much personal bias into which data you toss out and which data you accept at face value. 

That's generally why I like to focus on portfolio consistency across all start years rather than fixate on specific timeframes.  Whether you're looking at gold, stocks, or tulips, that kind of context helps weed out the obvious outliers. 
« Last Edit: August 23, 2017, 04:55:48 PM by Tyler »

AlmstRtrd

  • Stubble
  • **
  • Posts: 183
Re: Gold price and the hazards of backtesting
« Reply #7 on: August 23, 2017, 01:14:51 PM »
That's some nice research, Radagast. I like the CPI-U divisor. Will have to ponder your post a bit more.

To Tyler's credit I think he provides plenty of tools at portfoliocharts to allow investors to disregard the outperformance of gold in the early 1970s. For example, (and maybe there is just a basic idea I am missing) why can't one just look at the GB or PP heat maps and ignore whatever years one cares to ignore? Both of those have turned out to be excellent portfolios even when stripping away the early returns for gold.


effigy98

  • Bristles
  • ***
  • Posts: 369
Re: Gold price and the hazards of backtesting
« Reply #8 on: August 23, 2017, 02:13:37 PM »
Gold still looks great from 2000 to now (basically my investment years). So if we can be biased some years, I choose the years I was able to invest in the market and I trust those years will be closer to reality then the crusty 70/80s for me.

Growth of 10k from 2000
- Golden Butterfly: $37,020 (low volatility, boring... least I can sleep at night).
- 60/40: 26,570 (low volatility, easy and passive with vbinx, gold haters will accept me).
- Permanent Portfolio: $25,308 (low volatility, boring, message boards are hostile, too much gold!).
- 100% stocks: $25,117 (Probably should have 10% of your portfolio in Pfizer from all the Xanax you will be consuming).

I like this website from a real retiree perspective, he has 2000 at the bottom. MPT looks awesome for all people, but lots of rebalance. Larry (no gold) and PP (gold) look nice as well. Even 100% TIPs did pretty good! Crazy.
http://www.retireearlyhomepage.com/reallife17.html


« Last Edit: August 23, 2017, 02:17:17 PM by effigy98 »

SeattleCPA

  • Handlebar Stache
  • *****
  • Posts: 1442
  • Age: 59
  • Location: Redmond, WA
    • Evergreen Small Business
Re: Gold price and the hazards of backtesting
« Reply #9 on: August 24, 2017, 08:07:30 AM »
Another way to say this same thing: I'd love to work with the SWR your site shows for a Swensen portfolio... but gosh I wish there was about 100 more years of data!

So do I!  :)  I'm always on the lookout for more data history and will be happy to add it as I find it.  BTW, you might check out the Withdrawal Rates FAQ if you haven't already.

Tyler, I had read earlier and enjoyed your withdrawal rates article. I was particularly interested in something you say in it--which is that less portfolio variability supports a higher safe withdrawal rate.

That statement makes intuitive sense to me... and it meshes with my understanding of modern portfolio theory's math and the "sequence of returns" risk stuff. Further, I look at the withdrawal rate charts for portfolios with less risk (a lower standard deviation) and more "MPT" baked into them... and they sure seem to show this visually.

E.g., if one looks at the safe withdrawal rate of the total stock market portfolio, your chart suggests a 30-year rate of around 4.5% (and shows CAGR as 8.2% and StDev as 17.2%):

https://portfoliocharts.com/portfolio/total-stock-market/

If I look at safe withdrawal rate of Swensen porfolio, your chart suggests a 30-year rate of slightly more than 5% ( and then CAGR 6.8% and StDev 11.4%):

https://portfoliocharts.com/portfolio/swensen-portfolio/

I want, desparately, to believe I can bump my SWR by dampening portfolio volatility! :-)

And examples like the one above mesh with your article. But I also feel like correlations are so, er, fluid, that as I think you have said more than once in posts across the web, one needs to exercise caution.

BTW, to bring this back to gold, I think the only reason gold is important (and I think for some it could be important) is because it'll dampen portfolio volatility and then bump your safe withdrawal rate.

P.S. In Berntstein's Rational Expectations (a great little book),he discusses how a modest allocation to gold nicely improves your portfolio for basically this exact reason.

wienerdog

  • Bristles
  • ***
  • Posts: 463
Re: Gold price and the hazards of backtesting
« Reply #10 on: August 24, 2017, 05:27:17 PM »
Nixon 46 years ago on the 15th of Aug made it policy to no longer accept US dollars for gold.  There started to be a run on gold and he was going to control it.  He also suspended wage and price increases for 90 days to try to control inflation.  Go listen to his speech he says it was to defend the dollar against speculation.  He claimed your dollar would be worth just as much tomorrow as it is today which was a lie.  Shortly after that in December the dollar was devalued as it was at the beginning of 1971 also.  This created a panic and a loss of confidence in the dollar.  Further devaluations hurt it more in 73.

It started to unravel in the 60s as gold was pegged at $35/oz since 1934 by the treasury. In 65 many countries stopped using real silver in their coins because the value of silver started to be worth more than the monetary value of the coin. If you have a 64 dime it is worth $3 today.  If you have a 65 it is worth $0.10. 

You say look at the chart and gold was low and stable up to 72.  That is hardly the case if you study the 60s and early 70s closely.  By the late 60s the two tier market for gold was already agreed upon so you could go sell gold on the free market but countries could only pay country debt in gold at $35.  If you look at gold prices through the 60s it was hardly worth the exact amount of $35 that they claim they pegged it to.  In fact it is very hard to control.  By 69 it was worth $43 and in 70 it was worth less than $35.  Countries began to demand gold when they knew the value was up in the 40s.  Demand it at $35 and sell it at $40 on the London open market.  The price was driven all by US monetary expansion which ended up the US couldn't cover at $35.  Too much US dollars around the world.  Things like this and more US involvement in Vietnam, race riots of the early 70s, bombing in 71 by the Weather Underground and general instability during these times made them shut down the dollar for gold as a run had began.   They couldn't control the price like they thought they could. 

Maybe I am misunderstanding you but when you say "Based strictly on the history of gold prices, there is no reason to believe this sequence resulted from an external economic situation and is thus repeatable."  The thing you are missing is all it takes is for a loss in confidence of the dollar which surely is repeatable.  I think if you look at the M1 money supply and the slope we are on after 2009 you will see a distinct pattern that shows we are on the same path as back then.  All it will take is a loss in confidence of the dollar and the IMF will have to step in again.  I bet the price of gold isn't what it is now when it happens and I bet stock prices won't be either.

Car Jack

  • Handlebar Stache
  • *****
  • Posts: 1262
Re: Gold price and the hazards of backtesting
« Reply #11 on: August 25, 2017, 09:50:14 AM »
A couple of small, but very important points.

There was no investing in gold in the US before January 1, 1975.  Holding gold wasn't legal.

A 64 dime is not worth $3.  If you think it is and you're willing to back this up, I have probably $30 face value and will gladly sell it to you.  Apmex will sell you that dime for $1.36.

Tyler

  • Handlebar Stache
  • *****
  • Posts: 1158
Re: Gold price and the hazards of backtesting
« Reply #12 on: August 25, 2017, 10:02:59 AM »
There was no investing in gold in the US before January 1, 1975.  Holding gold wasn't legal.

This is a common misconception.  Private ownership of physical gold bullion was illegal in the US prior to 1975, but holding gold certificates (similar to the way modern gold ETFs work) was legalized in 1964.  https://en.wikipedia.org/wiki/Gold_Reserve_Act 
« Last Edit: August 25, 2017, 02:12:49 PM by Tyler »

wienerdog

  • Bristles
  • ***
  • Posts: 463
Re: Gold price and the hazards of backtesting
« Reply #13 on: August 25, 2017, 01:54:54 PM »

A 64 dime is not worth $3.  If you think it is and you're willing to back this up, I have probably $30 face value and will gladly sell it to you.  Apmex will sell you that dime for $1.36.

Uncirculated?  Sure I'll take you up on that.

Radagast

  • Handlebar Stache
  • *****
  • Posts: 1577
  • Location: West of the Mountains, East of the Sea
  • One does not simply work into Mordor
Re: Gold price and the hazards of backtesting
« Reply #14 on: August 27, 2017, 06:59:55 PM »
Sorry, I planned to discuss this topic but have been busy since I posted.

Radagast

  • Handlebar Stache
  • *****
  • Posts: 1577
  • Location: West of the Mountains, East of the Sea
  • One does not simply work into Mordor
Re: Gold price and the hazards of backtesting
« Reply #15 on: August 27, 2017, 07:11:29 PM »
Great post. I've said previously on this forum that people that use these asset allocation tools are probably doing themselves a disservice. The trick is to realise that the past had unique opportunities that you probably couldn't even obtain and that probably won't occur in the future. Therefore don't try and come up with the perfect asset allocation based on historical data.

I honestly think the smartest asset allocation is own your own home and invest in international stocks and domestic bonds as per your risk profile. It's the smartest because it's the simplest as well. I don't though follow my own advice.
I am not against back testing in general, I think most people take it with the big grains of salt it deserves. In this case though I think  that I would recommend against all back testing whatsoever of gold until 1974. If you are counting on market returns for your future finances, then back testing a period of government price manipulation is not indicative of what the market is likely to do.

Actually on of the best lessons I have learned from Portfoliocharts is that naive diversification works. Try typing a 16 or a 17 into every 4th box in the FIRE calculator https://portfoliocharts.com/portfolio/financial-independence/, then shift the numbers down one cell and try again until you tried them all. You can also try other equal weighting schemes (there are 24 boxes). The surprising thing is that pretty much every naive equally weighted combination historically supported a higher withdrawal rate than any permutation of a 2- or 3-fund portfolio. You also get similar results with Portfoliovisualizer, but sometimes you need to go through contortions to extend the data back enough to be meaningful.

Radagast

  • Handlebar Stache
  • *****
  • Posts: 1577
  • Location: West of the Mountains, East of the Sea
  • One does not simply work into Mordor
Re: Gold price and the hazards of backtesting
« Reply #16 on: August 27, 2017, 07:35:49 PM »
Back tests of gold beginning in early 1973, 1972, 1971, 1970, or earlier should not be regarded as valid. This is especially a critique of Portfoliocharts which presents many years of data into a single display. That in itself is not a problem, however, Portfoliocharts tends to emphasize the worst case returns, all of which would have been greatly mitigated by owning gold before the market could set its price, and rebalancing after it had done so. Many Portfoliocharts calculators also tend to obscure exactly which lines started in invalid years. Furthermore, it hatches out questionable years for many assets, but not for gold.

Thanks for your detailed and thoughtful gold explanation! 

I'll simply point out that the majority of the Portfolio Charts visuals are specifically designed to allow the user to see the big picture while discerning certain start years for any reason you like.  For the Heat Map, just ignore individual rows.  And for any chart with multiple lines, note that they are color coded by start date with the oldest lines darker than the newest ones.  The "hatching out" of data is solely related to assets where I've replaced the desired one that has no good data with an alternative similar asset, while in this case the data is fine and the issue is a matter of interpreting the numbers.  But now that you mention it, I can perhaps look into how to expand that functionality to isolate individual years for situations like this.  My goal is to provide as much good information as possible, not to obscure anything. 

Feel free to PM me if you have any suggestions.  I'm always open to new ideas.
I really like Portfoliocharts, its biggest asset is that it eliminates the cherry-picking that most people do when they back test. I did not mean that you intentionally obscure anything, just that by presenting all years equally at once it can be hard to discern which years are which results, which is generally a good thing.

In the case of gold though I think you should fade or hatch the data starting in 1970, 1971, 1972, and 1973 similar to how emerging markets are denoted in that period.  I have not read in depth in the history of this era, but skimming wikipedia it seems that gold was not completely set free to market prices until sometime in 1973. I know that it seems easy to include gold returns from this period because we great daily price data down to the penny from both London and the US, which is unusually detailed for this period. As I said in response to steveo, if you are counting on market returns for your future finances, then back testing a period of government price control is not indicative of what the market is likely to do in the future.

The problem is then whether I am cherry picking the start point. After further consideration I think annual prices from the beginning of 1974 on are probably "good enough". Actually my analysis pushed that to a year earlier than before, I used to think 1975 was the earliest reasonable start date but now I see that start date has its own cherry-picking issues. As a critique of my own OP, there were gold mines operating in the US continuously throughout history, including 1970. Compared to today they were small, mined shallow high-grade ore deposits, and were utterly free of health, safety, and environmental regulation. It it reasonable to assume that the cost to mine an ounce was lower than it is now.

Other real assets did well in this period in addition to gold. What would gold have done if its price had been free to move? What it actually did is the most optimistic answer. Would it have been more like oil? But oil went through its own artificial shortage, which was one of the causes of both inflation and the stock crash in '73-74. Would it have been more like REITs in your calculator? But those pay dividends while gold does not. The most pessimistic guess is that gold would have been similar to real estate sans cash flow. Without knowing how it would have performed in a free market, I think we should treat that period with skepticism.

Radagast

  • Handlebar Stache
  • *****
  • Posts: 1577
  • Location: West of the Mountains, East of the Sea
  • One does not simply work into Mordor
Re: Gold price and the hazards of backtesting
« Reply #17 on: August 27, 2017, 07:37:48 PM »
I've seen this info mentioned a few different times recently regarding those back-tested portfolios containing a gold component.
I had too, but they were all just guesses, wouldn't help you if they could. So I decided to look at actual numbers :-)

Radagast

  • Handlebar Stache
  • *****
  • Posts: 1577
  • Location: West of the Mountains, East of the Sea
  • One does not simply work into Mordor
Re: Gold price and the hazards of backtesting
« Reply #18 on: August 27, 2017, 07:48:14 PM »
But enough about Portfolio Charts -- Let's get back to Radagast's excellent gold discussion.

I'll certainly concede that looking at data alone often doesn't tell the whole story.  However, I don't think the problem is isolated to gold by any means.  For example, Robert Shiller argues in "Irrational Exuberance" that the US stock bubble in the 90's was just as much of a historical aberration as gold performance in the early 70's.  No backtesting tool will ever be a substitute for good sense, and I appreciate discussions like this that help put some of the numbers into proper historical context.  Just be careful not to add too much personal bias into which data you toss out and which data you accept at face value. 

That's generally why I like to focus on portfolio consistency across all start years rather than fixate on specific timeframes.  Whether you're looking at gold, stocks, or tulips, that kind of context helps weed out the obvious outliers.
I hope I am not not injecting too much bias into this. I am actually on the fence about gold as part of a portfolio, if I was against it I wouldn't have cared enough to put all this together. That being said, by the end of 1970 an ounce of gold was set at less than one consumer price index, which as I have shown, is far below any price the market has ever assigned it. Much as it would be nice to know what price the market would have assigned to gold throughout this period, I can't see that the actual prices assigned by the Nixon administration are a good indicator of this, nor the first few months after the market was allowed to determine the price.

Radagast

  • Handlebar Stache
  • *****
  • Posts: 1577
  • Location: West of the Mountains, East of the Sea
  • One does not simply work into Mordor
Re: Gold price and the hazards of backtesting
« Reply #19 on: August 27, 2017, 07:56:03 PM »
That's some nice research, Radagast. I like the CPI-U divisor. Will have to ponder your post a bit more.

To Tyler's credit I think he provides plenty of tools at portfoliocharts to allow investors to disregard the outperformance of gold in the early 1970s. For example, (and maybe there is just a basic idea I am missing) why can't one just look at the GB or PP heat maps and ignore whatever years one cares to ignore? Both of those have turned out to be excellent portfolios even when stripping away the early returns for gold.
The CPI is a convenience. It saves you the trouble of figuring out what year's currency value to use. As an example, suppose I wanted to update the figures in 2016. Would I still use 2016 currency? That would start to look out of date fast. So would I redo all the currency calculations? More work. Instead of dividing by one year's CPI and multiply by another, you can just divide once and leave it at that.

I am not knocking Portfoliocharts, which is a great tool. I am not even knocking gold, which is what it is. I am simply questioning whether the pricing of gold until 1974 or so might be unusually bad even by the questionable standards of back testing. I agree that the heat map makes it easier for a user to eliminate possible suspect years.

Radagast

  • Handlebar Stache
  • *****
  • Posts: 1577
  • Location: West of the Mountains, East of the Sea
  • One does not simply work into Mordor
Re: Gold price and the hazards of backtesting
« Reply #20 on: August 27, 2017, 07:59:04 PM »
Gold still looks great from 2000 to now (basically my investment years). So if we can be biased some years, I choose the years I was able to invest in the market and I trust those years will be closer to reality then the crusty 70/80s for me.

Growth of 10k from 2000
- Golden Butterfly: $37,020 (low volatility, boring... least I can sleep at night).
- 60/40: 26,570 (low volatility, easy and passive with vbinx, gold haters will accept me).
- Permanent Portfolio: $25,308 (low volatility, boring, message boards are hostile, too much gold!).
- 100% stocks: $25,117 (Probably should have 10% of your portfolio in Pfizer from all the Xanax you will be consuming).

I like this website from a real retiree perspective, he has 2000 at the bottom. MPT looks awesome for all people, but lots of rebalance. Larry (no gold) and PP (gold) look nice as well. Even 100% TIPs did pretty good! Crazy.
http://www.retireearlyhomepage.com/reallife17.html
Its history since 1985 demonstrates that it may have the potential to help in bad times, while dragging you down in good times, which would be a desirable property if it is reliable.

SeattleCPA

  • Handlebar Stache
  • *****
  • Posts: 1442
  • Age: 59
  • Location: Redmond, WA
    • Evergreen Small Business
Re: Gold price and the hazards of backtesting
« Reply #21 on: August 28, 2017, 07:05:44 AM »
... naive diversification works. Try typing a 16 or a 17 into every 4th box in the FIRE calculator https://portfoliocharts.com/portfolio/financial-independence/, then shift the numbers down one cell and try again until you tried them all. You can also try other equal weighting schemes (there are 24 boxes). The surprising thing is that pretty much every naive equally weighted combination historically supported a higher withdrawal rate ...

Radagast, I think your approach basically demonstrates how modern portfolio theory thinking (specifically, uncorrelated asset classes) improves portfolio construction.

As I noted above in thread, I would love to bank on the higher withdrawal rates supported by an MPT-savvy portfolio. (And I should say, in essence, I am banking on the higher withdrawal rates an MPT-ish portfolio supports.) But I think the two problems with assuming higher withdrawal rates are:

1. The asset class correlations change over time so you and I can't spend too much time tweaking the percentages. (I'm going to blog about this next week, but Bernstein notes in "Rational Expectations" that after you add riskless assets to your portfolio, throwing in additional uncorrelated asset classes gets pretty inefficient in terms of predictably dampening volatility...)

2. In a "black swan" event, nearly all asset classes may become highly corrrelated...

In the end, my own personal thinking is (1) construct a diversified portfolio and hope historical asset class correlations hint at future asset class correlations... (2) admit possibility of lower withdrawal rates than PortfolioCharts shows.

Tyler

  • Handlebar Stache
  • *****
  • Posts: 1158
Re: Gold price and the hazards of backtesting
« Reply #22 on: August 28, 2017, 10:01:33 AM »

1. The asset class correlations change over time so you and I can't spend too much time tweaking the percentages. (I'm going to blog about this next week, but Bernstein notes in "Rational Expectations" that after you add riskless assets to your portfolio, throwing in additional uncorrelated asset classes gets pretty inefficient in terms of predictably dampening volatility...)

2. In a "black swan" event, nearly all asset classes may become highly corrrelated...

In the end, my own personal thinking is (1) construct a diversified portfolio and hope historical asset class correlations hint at future asset class correlations... (2) admit possibility of lower withdrawal rates than PortfolioCharts shows.

Correlations do change and that's why it pays to have more than two assets.  But you're correct that there's also a practical limit to diversification.  IMHO, it's not about buying as many different index funds as possible but about wisely spreading your money among a few different major categories. 

Regarding withdrawal rates, the safest way to interpret the data is to use it to compare relative withdrawal rates between portfolios but to assume that the absolute numbers for all portfolios may be a little lower in real life (once you account for taxes, fees, data limitations, etc).  When dealing with your life savings, be smart about it and plan conservatively. 
« Last Edit: August 28, 2017, 10:28:21 AM by Tyler »

Car Jack

  • Handlebar Stache
  • *****
  • Posts: 1262
Re: Gold price and the hazards of backtesting
« Reply #23 on: August 28, 2017, 01:09:36 PM »

A 64 dime is not worth $3.  If you think it is and you're willing to back this up, I have probably $30 face value and will gladly sell it to you.  Apmex will sell you that dime for $1.36.

Uncirculated?  Sure I'll take you up on that.

Haha....no.  Junk silver.

Also remember that when it was 1964, you couldn't buy an uncirculated dime for a dime, right?  Just like today.

SeattleCPA

  • Handlebar Stache
  • *****
  • Posts: 1442
  • Age: 59
  • Location: Redmond, WA
    • Evergreen Small Business
Re: Gold price and the hazards of backtesting
« Reply #24 on: August 28, 2017, 04:36:37 PM »

1. The asset class correlations change over time so you and I can't spend too much time tweaking the percentages. (I'm going to blog about this next week, but Bernstein notes in "Rational Expectations" that after you add riskless assets to your portfolio, throwing in additional uncorrelated asset classes gets pretty inefficient in terms of predictably dampening volatility...)

2. In a "black swan" event, nearly all asset classes may become highly corrrelated...

In the end, my own personal thinking is (1) construct a diversified portfolio and hope historical asset class correlations hint at future asset class correlations... (2) admit possibility of lower withdrawal rates than PortfolioCharts shows.

Correlations do change and that's why it pays to have more than two assets.  But you're correct that there's also a practical limit to diversification.  IMHO, it's not about buying as many different index funds as possible but about wisely spreading your money among a few different major categories. 

Regarding withdrawal rates, the safest way to interpret the data is to use it to compare relative withdrawal rates between portfolios but to assume that the absolute numbers for all portfolios may be a little lower in real life (once you account for taxes, fees, data limitations, etc).  When dealing with your life savings, be smart about it and plan conservatively.

I totally, totally agree. And just to say this one more time, I love PortfolioCharts.com.

VoteCthulu

  • Bristles
  • ***
  • Posts: 409
Re: Gold price and the hazards of backtesting
« Reply #25 on: August 28, 2017, 09:09:42 PM »
That's generally why I like to focus on portfolio consistency across all start years rather than fixate on specific timeframes.  Whether you're looking at gold, stocks, or tulips, that kind of context helps weed out the obvious outliers.
Where is the chart that includes all of those years gold was essentially worth 0 in the US? Does any portfolio with gold weather that time well?

Radagast

  • Handlebar Stache
  • *****
  • Posts: 1577
  • Location: West of the Mountains, East of the Sea
  • One does not simply work into Mordor
Re: Gold price and the hazards of backtesting
« Reply #26 on: August 28, 2017, 11:17:39 PM »
You say look at the chart and gold was low and stable up to 72.  That is hardly the case if you study the 60s and early 70s closely.  By the late 60s the two tier market for gold was already agreed upon so you could go sell gold on the free market but countries could only pay country debt in gold at $35.  If you look at gold prices through the 60s it was hardly worth the exact amount of $35 that they claim they pegged it to.  In fact it is very hard to control.  By 69 it was worth $43 and in 70 it was worth less than $35.  Countries began to demand gold when they knew the value was up in the 40s.  Demand it at $35 and sell it at $40 on the London open market.  The price was driven all by US monetary expansion which ended up the US couldn't cover at $35.  Too much US dollars around the world.  Things like this and more US involvement in Vietnam, race riots of the early 70s, bombing in 71 by the Weather Underground and general instability during these times made them shut down the dollar for gold as a run had began.   They couldn't control the price like they thought they could.
The charts I showed in OP are for US gold exchange, but I first looked at London spot prices thinking along those lines. Here is a graph comparing US and London gold prices. Note that these prices are the closing price of the last day of the year, so there is more noise than you might expect (I think many places which report historic gold prices use various length averages or something). I think a good analogy is that US and UK gold prices track each other like an illiquid ETF tracks its index.

Orange is London and grey is US. Did you notice how smooth and stable the price was from 1955 until 1970?

But that does bring up another point I have. It is not only the low price, it is also the timing of gold price fluctuations which is suspicious. By 1970 inflation, political uncertainty, and general fiscal instability were all well underway. If gold is supposed to go up in response to those, it should not have been at an all time low in 1970. I take this as another indication that gold prices and returns in these years are in no way a reliable test of what gold should do in open trading.
Quote
Maybe I am misunderstanding you but when you say "Based strictly on the history of gold prices, there is no reason to believe this sequence resulted from an external economic situation and is thus repeatable."  The thing you are missing is all it takes is for a loss in confidence of the dollar which surely is repeatable.  I think if you look at the M1 money supply and the slope we are on after 2009 you will see a distinct pattern that shows we are on the same path as back then.  All it will take is a loss in confidence of the dollar and the IMF will have to step in again.  I bet the price of gold isn't what it is now when it happens and I bet stock prices won't be either.
It was caused by an artificially low fixed price, followed by an adjustment to a reasonable market price. It did not go up because stocks and bonds were going down. It went up to correct an obvious mispricing in real terms. For this reason, you can't learn anything about subsequent history from back testing gold through this period (I suppose you could guess that anything under 1 CPI is too low, if you observe it in the future).  The price on anything could go up quickly for various reasons.

Radagast

  • Handlebar Stache
  • *****
  • Posts: 1577
  • Location: West of the Mountains, East of the Sea
  • One does not simply work into Mordor
Re: Gold price and the hazards of backtesting
« Reply #27 on: August 28, 2017, 11:29:29 PM »
... naive diversification works. Try typing a 16 or a 17 into every 4th box in the FIRE calculator https://portfoliocharts.com/portfolio/financial-independence/, then shift the numbers down one cell and try again until you tried them all. You can also try other equal weighting schemes (there are 24 boxes). The surprising thing is that pretty much every naive equally weighted combination historically supported a higher withdrawal rate ...

Radagast, I think your approach basically demonstrates how modern portfolio theory thinking (specifically, uncorrelated asset classes) improves portfolio construction.

As I noted above in thread, I would love to bank on the higher withdrawal rates supported by an MPT-savvy portfolio. (And I should say, in essence, I am banking on the higher withdrawal rates an MPT-ish portfolio supports.) But I think the two problems with assuming higher withdrawal rates are:

1. The asset class correlations change over time so you and I can't spend too much time tweaking the percentages. (I'm going to blog about this next week, but Bernstein notes in "Rational Expectations" that after you add riskless assets to your portfolio, throwing in additional uncorrelated asset classes gets pretty inefficient in terms of predictably dampening volatility...)

2. In a "black swan" event, nearly all asset classes may become highly corrrelated...

In the end, my own personal thinking is (1) construct a diversified portfolio and hope historical asset class correlations hint at future asset class correlations... (2) admit possibility of lower withdrawal rates than PortfolioCharts shows.
I agree I wouldn't use this as a strategy to attempt a higher SWR than I would otherwise. Instead I would tend to view it as a way to make that SWR more reliable. Another way of saying this is that a 5% withdrawal rate from 100% Total US Stock Market has succeeded most of the time. Adding in uncorrelated assets would not increase the SWR beyond 5%, but it should make 5% more likely to succeed (or 4%, or whatever your number is).

Of course another possibility is that each of those alternatives is simply a list of "things that did OK in the '70's" :). But then, the same principle also seems to have held up in the 2000's.

SeattleCPA

  • Handlebar Stache
  • *****
  • Posts: 1442
  • Age: 59
  • Location: Redmond, WA
    • Evergreen Small Business
Re: Gold price and the hazards of backtesting
« Reply #28 on: August 29, 2017, 07:55:31 AM »
I agree I wouldn't use this as a strategy to attempt a higher SWR than I would otherwise. Instead I would tend to view it as a way to make that SWR more reliable. Another way of saying this is that a 5% withdrawal rate from 100% Total US Stock Market has succeeded most of the time. Adding in uncorrelated assets would not increase the SWR beyond 5%, but it should make 5% more likely to succeed (or 4%, or whatever your number is).

Of course another possibility is that each of those alternatives is simply a list of "things that did OK in the '70's" :). But then, the same principle also seems to have held up in the 2000's.

I think I'm on the same page.

It seems like trying to construct a portfolio with more uncorrelated assets will help with things like SWR. But one can't bank on that. But it'll probably help. Maybe. Or Maybe not. Then again, why not try to do this. It isn't that hard. Etc., etc.

wienerdog

  • Bristles
  • ***
  • Posts: 463
Re: Gold price and the hazards of backtesting
« Reply #29 on: August 29, 2017, 10:44:47 AM »
Also remember that when it was 1964, you couldn't buy an uncirculated dime for a dime, right?  Just like today.

True

FrugalFisherman10

  • Bristles
  • ***
  • Posts: 310
    • Fly Fishing Photo Project
Re: Gold price and the hazards of backtesting
« Reply #30 on: August 30, 2017, 07:21:17 AM »
Very timely post Radagast! I have been looking into all this for a little while now and conversing with Tyler about how to best use his charts as well. So, many thanks for laying this all out there to consider.
And a very public thank you to Tyler who keeps pushing the envelope on visualization... on his own time. your work is very helpful and is a great teacher.

Much as it would be nice to know what price the market would have assigned to gold throughout this period, I can't see that the actual prices assigned by the Nixon administration are a good indicator of this, nor the first few months after the market was allowed to determine the price.

Now that you have figured out what price the market assigned to gold (within a few months of the market being allowed to do so), would it be too artificial of a proxy to use that price for 'all time leading up to 1973'? Just an idea. That would smooth things out quite a bit, but would nonetheless be pretty 'artificial' / unreliable.


1. The asset class correlations change over time so you and I can't spend too much time tweaking the percentages. (I'm going to blog about this next week, but Bernstein notes in "Rational Expectations" that after you add riskless assets to your portfolio, throwing in additional uncorrelated asset classes gets pretty inefficient in terms of predictably dampening volatility...)

Correlations do change
and that's why it pays to have more than two assets.  But you're correct that there's also a practical limit to diversification.  IMHO, it's not about buying as many different index funds as possible but about wisely spreading your money among a few different major categories. 


Excerpts: The Volatility of Correlation: Important Implications for the Asset Allocation Decision by William J. Coaker II,* The average variance in correlation measured 0.98 over one year and 0.25 over ten years. In short, the relationship among many of the asset classes appears to be inherently unstable.
* Correlations exhibit uniqueness, meaning periods are distinct from previous time periods. For example, international stocks' correlation to the S&P 500 was 0.48 from 1970 to 1997, but 0.83 from 1998 to 2002.
* Rather than rely on historical correlations, a more comprehensive and dynamic approach is needed in making asset allocation decisions.   
WTF kind of cliff hanger is that?
 
...Here’s his next paper:

Emphasizing Low-Correlated Assets: The Volatility of Correlation by William J. Coaker II
• The fact that correlations change is well known. But the severity of change, and which relationships are subject to change, needs to be better understood because it has important implications for containing risk.
• This study evaluates the volatility of correlation among 18 asset classes to each other to determine the consistency or inconsistency of relationships.
• In the asset allocation process, some assets often are used together even though diversification benefits have been very low. (high correlation = low diversification benefits) For example, the correlations of the S&P 500 to large growth, mid-blend to mid-growth, small blend to small growth, and large value to mid-value, have been very strong.
• Several assets often are neglected in the asset allocation decision, even
though their diversification benefits have been very high. Natural resources, global bonds, and long-short, for example, stand out as having consistently low correlations to all the other assets in this study.
Growth and blend styles are highly correlated, and using them together does little to reduce risk.
Real estate, high-yield bonds, U.S. bonds, and long-short are more closely linked to value investing than growth. Emerging markets are somewhat more connected to growth than value.

• The asset allocation decision should emphasize low-correlated assets that satisfy return objectives. Two sample portfolios for different style investors show how risk and return are improved by combining lower-correlated assets
https://forum.mrmoneymustache.com/investor-alley/what-am-i-to-make-of-this/msg1331388/#msg1331388

SeattleCPA

  • Handlebar Stache
  • *****
  • Posts: 1442
  • Age: 59
  • Location: Redmond, WA
    • Evergreen Small Business
Re: Gold price and the hazards of backtesting
« Reply #31 on: August 30, 2017, 10:11:04 AM »
Frugalfisherman, good stuff there. Thanks for posting.

I find myself thinking this is all a little like Pascal's Wager...

Wikipedia's description: https://en.wikipedia.org/wiki/Pascal%27s_Wager

The paraphrased for application to MPT: As long as you don't spend too much money and bet too heavily, why not try to build a portfolio with uncorrelated assets.

Tyler

  • Handlebar Stache
  • *****
  • Posts: 1158
Re: Gold price and the hazards of backtesting
« Reply #32 on: August 30, 2017, 06:16:39 PM »
The charts I showed in OP are for US gold exchange, but I first looked at London spot prices thinking along those lines. Here is a graph comparing US and London gold prices. Note that these prices are the closing price of the last day of the year, so there is more noise than you might expect (I think many places which report historic gold prices use various length averages or something). I think a good analogy is that US and UK gold prices track each other like an illiquid ETF tracks its index.

Orange is London and grey is US. Did you notice how smooth and stable the price was from 1955 until 1970?

Great chart. 

Regarding the difference between London and US gold prices, I learned in my own research that at least part of the discrepancy is simply due to differences in sampling times.  They're never going to be exactly the same because the data is measured at different times of the day.

I'd also like to point out that this chart provides some nice context for the gold rise from 1971-1974 due to the end of the gold standard.  You can see in the chart that we're just talking about the first small peak on the left.  From the way some people talk about that initial gold correction after the price fix was removed, I think a lot of casual readers picture that peak to be way higher and assume it must dwarf all gold returns after that timeframe.  But the subsequent spikes around 1980 and 2010 were no longer affected by that initial price correction, and that kind of normal volatility is a good visual of how gold works in a fiat currency world.  So feel free to (rightly) look at the years prior to 1975 with a skeptical eye, but be sure to appreciate how it performed in the years after that as well. 

And just so there's no misunderstanding, I'm not at all trying to push gold on anyone as there are plenty of good portfolios that don't require it.  Just offering a little context for those looking to understand how it works. 
« Last Edit: August 31, 2017, 07:35:36 AM by Tyler »