Let’s take a look over the past year first. (See the Market Analysis page for sequential comparisons.) The first three quarters of 2022 were notably bullish for cash, in particular the US dollar. Synthetic Systems however only hinted at that at the beginning of the year … it consistently underestimated the strength of the USD relative to other assets. My call for cash outperformance on January 18 and at several subsequent points was based at least as much on subjective impressions of the state of financial markets, consumer price inflation, and the outlook for Fed policy response. This call was in effect until I withdrew it on November 11, having concluded on October 27 that the bottom for bonds was likely in.
The value added by SS was minimal in the second half of the year. It bailed on stocks at the beginning of the third quarter as they continued to rally into December. This is not unusual; its resolution dissipates below about one quarter of a year. It did however help us identify the bottom in bonds … SS pegged it for October 1, and it came October 24. The likelihood that a low was due in that time frame helped us recognize it much sooner after it occurred than otherwise be the case.
Rather the major contribution of SS came in the first half of the year. Financology went deeply underweight bonds at the beginning of the year and remained so until returning to neutral around mid year, finally going overweight at the beginning of the fourth quarter.
The overarching theme for the SS 2023 forecast is for Bonds and Gold to outperform and Stocks to underperform. Bills have a moderately bearish outlook but given the failure of SS to clearly call the bull market of 2022 it’s not a high confidence call. Copper has a moderately bearish outlook as well but has been unstable between successive forecasts and so I likewise hesitate to draw any firm conclusions there.
About Synthetic Systems
Synthetic Systems is a computer forecasting model covering five asset classes: Treasury bills, Treasury bonds, stocks, copper & gold. The plots are all on the same basis so as to be directly comparable – total return. So the slope of one asset rising more than another indicates outperformance, and vice versa. Plots are in natural log space so that the same vertical increment means the same proportional (percentage) increase regardless of vertical position. The total returns are not denominated in dollars or any other currency, but are relative to each other.
Bills refers to US Treasury securities of effectively zero maturity and duration, similar to a Treasury money market or ultrashort bond fund. Bonds refers to US Treasury securities of effectively infinite maturity (duration reciprocal of yield), but approximates the returns of real world extended duration Treasury funds like EDV, and to lesser extent VGLT and TLT. A broad UST fund like GOVT is around midway between Bills and Bonds. Stocks refers to the entire asset class (not just US or any one country). Copper and Gold represent the respective elementary commodities, except to note that copper is broadly indicative of physical commodities as a class and suggestive of trends in goods and services price inflation.
Because the Bonds plot represents infinite maturity bonds, it is greatly exaggerated in relation to most real-world bond investments. The Vanguard extended duration Treasury fund EDV tracks it closely. Other popular long term treasury funds such as TLT and VGLT can be interpolated as a roughly 90:10 mix of Bonds and Bills. The broad iShares treasury fund GOVT is roughly a 40:60 mix.
The choice of modeling infinite maturity bonds may seem odd, but was motivated by a desire to put long bonds on an equal footing with stocks, which are infinite maturity securities. It does however ask the viewer to interpolate between the Bills and Bonds plots to visualize the performance of real world treasury investments. But this is true of many real world assets … junk bonds for instance can be read as intermediate bonds and stocks, and silver as intermediate copper and gold. SS asset choices are intended to represent the ‘corners’ of a region in which most real world assets reside.
Another unrelated factor could exaggerate the bond performance forecast. SS frequency response focuses on time frames from about one quarter year to four years. This means that it does not attempt to take into account very short or very long term trends. On the second count, if the secular trend of bond prices has turned lower, it won’t much be reflected in the forecast. A multi-decade trend would be overwhelmed by the cyclical trends SS focuses on, so it wouldn’t invalidate the bullish bond forecast over just a couple years, but it would reasonably be expected to moderate it.
At the other end of the spectrum, if forecast accuracy fades as the time frame grows shorter, why don’t I just filter that information out of the plots? The answer is that it’s still better than random. Even odds of 51%-49% on a forecast have utility, even if less so than trends in SS’s sweet spot. The practical impact is that readers of SS charts need to be aware of the time frame factor. It’s more work for the reader, but the cost of throwing out information of even modest marginal utility is difficult to justify.
The charts are best considered together. Annual and quarterly charts are respectively grouped together on dedicated pages under Market Analysis to facilitate this. The forecasting accuracy of Synthetic Systems is best evaluated by comparing successive charts, as the “Projected” time frame of an earlier chart slides to the left into the “Actual” time frame of later charts. Moreover, similarities between the latest update and earlier updates indicate areas of greater confidence in contrast to differences which indicate greater uncertainty.
Readers should bear in mind that Synthetic Systems forecasts comprehensively reflect financial and economic forces (e.g inflation, interest rates, monetary policy, money flows, seasonality, natural resources, technology, demographics, the business cycle, global economic trends, consumer sentiment, investor psychology, momentum, mean reversion, etcetera), but do not reflect external noneconomic factors (e.g. natural disasters, pandemics, unexpected geopolitical disruptions) until they are incorporated into the financial and economic sphere. It’s most applicable over time frames from one quarter year to four years … its accuracy is limited by noise and news flow on the shorter time frames and it also does not attempt to model drivers of longer term returns such as valuations.
Readers are encouraged to consider Synthetic Systems forecasts in conjunction with fundamentals and valuations. Synthetic Systems however often forecasts trends long before the fundamentals fall into place; for this reason it’s useful as a medium term planning guide, with the appearance of fundamentals consistent with a Synthetic Systems forecast serving as confirmation, the lack thereof nonconfirmation. Final confirmation occurs when the trend is actually in evidence.