Asset Allocation Beyond the Zero Bound

Japan has had zero rates for 20 years. Here is what seems to have mattered for investment outcomes.
 

By: Verdad Research

We appear to be entering a period of prolonged low interest rates across most major developed markets. US and European investors are wrestling with how to adapt to this “new normal”—and how their asset allocation decisions should shift in a zero-rate environment.

Japanese investors, however, have dealt with these conditions for over 20 years since the initiation of the Zero Interest Rate Policy (ZIRP). Japan dropped rates to zero in February of 1999, and Japanese government bonds have been at around a 1% yield ever since. Zero rates aren’t a new normal; they’re just normal in Japan.

Japan thus provides an interesting case study for how different assets performed in a zero-rate environment and which asset allocation decisions made the most sense in our view. Below we show Japanese domestic asset class returns since ZIRP began.

Figure 1: Japanese Domestic Asset Class Returns Post ZIRP (2/26/1999 to 2/26/2020)

Source: MSCI and Bloomberg. MSCI Factor indexes for stocks, Bloomberg Barclays Japan Treasury Index and S&P Japan Corporate Bond Index for bonds, and MSCI commodity producers and REIT indexes as rough proxies for commodities and real estate. Annualized total returns in Yen.

In terms of absolute performance, small value stocks and real estate were the best performing investments. Large growth stocks and commodities were the worst performing investments. Bonds, large value stocks, and small growth stocks performed in between.

The outperformance of small value stocks and real estate was a sharp reversal of prior trends. Both had significantly lagged the equity market in the 1990s. Small value stocks were out of favor as investors chased large growth stocks during the technology bubble, and real estate was still reeling from the major collapse in real estate prices that began in 1990.

Despite major concerns over Japan’s debt/GDP levels and bond yields starting at 1.4%, Japanese bonds returned 1–2% over the period with almost no volatility, negative correlation to equities, and a Sharpe Ratio greater than one. Commodities, the only asset which does not produce cash flow, returned basically zero real returns over this period during which there was no inflation in Japan.

Only bonds and foreign exchange markets offered meaningful diversification benefits to equities, with real estate and commodities returns highly correlated to the stock market.

Figure 2: Correlation Matrix (Domestic, Yen Denominated)

Source: Capital IQ. For simplicity, small value used for stocks, and JGBs used for bonds.

In a zero-rate environment, it would have been easy to declare the death of bonds as a diversifying tool. But while measly yields correctly portended low returns from bonds, they did not mean that bonds lost their role as effective diversifiers. In fact, the negative correlation of Japanese government bonds with Japanese equities was stronger than the same relationship in the United States over this period. We believe owning bonds in a portfolio would have significantly reduced drawdowns and volatility in the Japanese example, something to keep in mind as we now read hot takes about the death of the 60/40 portfolio.

The performance of Japanese banks was also quite interesting. As shown below, Japanese bank stocks did horribly, as banks found themselves stuck with massive portfolios of zombie loans at the start of the period and were unable to make much money in a zero-rate environment.

Figure 3: Stock Returns of Japanese Banks vs. Equity Indices Since 1995

Source: Capital IQ

What’s perhaps most interesting about the performance of these different asset classes is that they were nearly perfectly predicted by the yields of the different asset classes in 1999. Below are the cash yields on each asset class at the start of the period and the subsequent 20-year returns.

Figure 4: Cash Yields in 1999 and Subsequent 20-Year Annualized Returns

Source: Capital IQ. Stock yields in Feb 1999 are calculated from the unlevered FCF yields on all listed stocks above/below $2bn in market cap for large/small stocks respectively. Value is the 20th and growth the 80th percentile breakpoint within those two universes. Bonds use the bond yield. Real estate yield is the Japanese cap rate from ARES, which is about the same as the J-REIT cap rate from Sumitomo Mitsui Trust Research Institute.

In retrospect, with basically zero GDP growth and zero inflation, we might expect entry yields to dominate outcomes over such long horizons. And it is remarkable to us how highly correlated the returns were with the yields at entry. Apart from Japanese banks and Japanese bond returns, there was no clear relationship between any of the asset classes’ relative performance that we could attribute to the prolonged zero-interest environment. Instead, entry valuations were highly predictive of actual outcomes in our view.

At the start of the period, bonds were very expensive at 1.4% yields. Real estate had become much cheaper, which is an understatement coming out of the 1989 asset bubble, when the small plot of land around the Imperial Palace in Tokyo was famously estimated to be worth more than all of the land in the state of California. Value and growth stocks were pushed to opposite historical extremes of the valuation spectrum at the end of the 2000 dot-com bubble. Some things were historically very expensive, and some things were historically pretty cheap in 1999. Lower expectations seemed to lead to higher returns on long horizons.

Many investors rely on historical returns and correlations of broad asset classes to determine asset allocation and form the basis for estimating expected future returns. But researchers have found and reaffirmed that, while practitioners typically use only 60 to 120 months of data to estimate optimal portfolio allocations, in reality, one would need 3,000 months to draw valid statistical inferences on 25 different assets and 6,000 months to draw valid statistical inferences with 50 assets to choose from. “The severity of estimation error is startling,” they conclude. However, they also found that “using information about the cross-sectional characteristics of assets, rather than just statistical information about the moments of asset returns, does lead to an improvement in Sharpe ratios.”

Perhaps the most significant conclusion to draw from Japan’s experience during ZIRP therefore is not about the special impact of a low-rate environment. As we see it, interest rates were relevant in predicting bond returns and the poor performance of bank stocks, but other asset classes were better predicted by relying on entry valuations and yields.