In shifting responsibility to Clough, Citron mentioned that he had relied on Clough’s 1993 prediction of flat or falling interest rates for three to five years. Even after the Federal Reserve boosted short-term rates in February 1994, Mr. Citron said Mr. Clough told him at a breakfast meeting on March 1 that rate increases were “not sustainable.”
However, starting in 1992 officials from Merrill’s risk management desk had been urging Citron to lower his leverage. They had even offered to repurchase some of the inverse floaters at a time when these would have produced a profit for the county. Moreover, in February 1994 they warned Citron that the interest rate outlook was uncertain. But Citron disregarded the warning from the risk management group, which conflicted with his own outlook. Instead, he chose Clough’s forecast, which confirmed that outlook. He paid attention to confirming evidence while ignoring disconfirming evidence, thereby succumbing to the illusion of validity. When faced with cognitive dissonance, most people resolve the dissonance by choosing the comfortable route.
Subsequently, Orange County played the blame game too, suing a number of parties with whom it had had a relationship. The most notable was Merrill Lynch, against whom it filed both criminal and civil charges, arguing that the firm had provided faulty advice. Both suits were settled. In 1997, Merrill Lynch agreed to pay $30 million to have the criminal suit dropped. In June 1998, it agreed to pay $400 million to have the civil suit dropped, the second largest suit ever involving a brokerage firm.
Orange County also filed suit against the bond-rating agency Standard & Poor’s Corp. for having given its top rating to the county. This raises the question of whether or not agencies incur legal liability for misjudgment, an interesting question indeed given the propensity for behavioral biases and errors.
The range of lawsuits filed by Orange County raises the general issue of fairness. Events in Orange County featured the misjudgment of Robert Citron, the aggressive sales tactics of some Merrill Lynch employees, the cautious warnings by other Merrill Lynch employees, and the mistaken opinions of bond-rating agencies. Is there a fair way to share the blame? The blame game is a fairness issue with which the investment community repeatedly struggles.
An interesting article appeared in the summer issue of the Journal of Derivatives. The article was written by Merton Miller and David Ross (1997) and titled “The Orange County Bankruptcy and Its Aftermath: Some New Evidence.” Miller and Ross argue that it was not necessary for Orange County to have declared bankruptcy, since in December 1994 the Orange County Investment Pool was neither insolvent nor illiquid. Moreover, they point out that had the portfolio simply been held rather than liquidated, subsequent interest rate moves would actually have led to a profit. The county would have emerged ahead, recovering the full $21 billion of its portfolio plus $300 million of interest.18
Appropriate actions are always clear in hindsight. That’s why the phenomenon is called hindsight bias. But Miller and Ross would not accept hindsight bias as a reason. Rather they contend that “the positive cash flows earned in 1995 are fully consistent with expectations based on historical term structure patterns.” Or, to put it somewhat differently, they suggest that interest rates are predictable.
Indeed, a December 30, 1994, article that appeared in Investor’s Business Daily concerning the Orange County fiasco stated: “There is still an incentive to leverage assets at the front end of the yield curve, where very short-term rates are at least 200 basis points lower than two-year rates.”19
We move now from a discussion about the specific events surrounding the Orange County bankruptcy and the predictability of interest rates in December 1994 to a discussion about the general predictability of interest rates.
The Expectations Hypothesis of the Yield Curve
In his 1995 survey article, John Campbell describes some of the main features associated with yield curves. The most well-known theoretical property is the expectations hypothesis. This hypothesis concerns the question of whether future interest rate movements can be predicted based on the yield spread.
To explain the expectations hypothesis, consider a situation like the one that prevailed in 1993 when long-term rates were higher than short-term rates. Look at the right side of figure 14-2. If investors expected the yield curve to remain unchanged through time, they would have an incentive to do what Robert Citron did: Sell the short-term securities short and use the proceeds to purchase the long-term securities.
If all investors thought along these lines, then they would all be following what I’ll call a “Citron strategy.” So, what expectations would prevent them from behaving in that manner? For one, investors would have to believe that the yield curve would move. But in what direction? Consider an investor with a short time horizon. What kind of expectations would this investor have to hold in order to think twice about a Citron strategy? One possibility is that he would expect long-term yields to rise: The expected capital loss on his long-term securities would provide the disincentive. Another possibility is that although he does not expect long term yields to rise, he perceives the risk that they might as a sufficient deterrent. In other words, he perceives the term premium as insufficient compensation for bearing this risk.
How about an investor with a long time horizon? If she were to think twice about a Citron strategy then she would be expecting short-term rates to rise enough to offset any advantage to the current spread.
Putting the last two points together leads to the following implications. An upward-sloping yield curve reflects investors’ expectations that both short-term rates and long-term rates will rise, but long-term rates will rise more. Moreover, the expected return difference between holding long-term bonds and holding three-month Treasury bills would have to be zero. This implication is testable.
The evidence reported by Campbell in connection with this implication is portrayed in figure 14-3, for the period 1952–1991. The upward-sloping curve shows the average yield spread between Treasurys of various maturities and the three-month Treasury bill. The higher the maturity, the greater the spread. The second curve shows how the average holding return varied across different maturities relative to the yield on three-month Treasury bills. According to the expectations hypothesis, the excess return should be zero for all maturities. But in reality, the excess return was positive for most maturities, declining with longer maturities. For the ten-year maturity, the excess return was actually negative.
Campbell also reports that wider spreads tended to be followed by long-term yields moving down rather than up, as the expectations hypothesis predicts. In this respect, he offers the following comment:
This is exactly the behavior noted by Macaulay (1938) in a classic work on the movements of asset prices in the late nineteenth and early twentieth centuries. Macaulay wrote (p. 33): “The yields of bonds are higher if the highest grade should fall during a period in which short-term rates are higher than the yields of the bonds and rise during a period in which the short-term rates are lower. Now experience is more nearly the opposite.” It is particularly impressive that this finding in the late nineteenth century appears in the late twentieth century as well as in Macaulay’s data. (p. 139)
In order to understand why the expectations hypothesis fails, we need to examine investors’ expectations, not just the subsequent behavior
Figure 14-3 Evidence about the Expectations Hypothesis of the Term Structure
The upward sloping curve in this figure shows the average spread between the Treasury yield for a particular maturity and the three-month Treasury bill, for the period 1952–1991. According to the expectations hypothesis of the yield curve, the excess return to holding a Treasury security of a given maturity, relative to Treasury bills, should be zero across all maturities. But the pattern is not flat. It rises and then falls. Indeed, the return to the ten-year bond was negative over this period. of yields of various maturities. Kenneth Froot (1989) undertook the first study that used investors’ interest rate forecasts to examine thi
s issue. He used data from a survey conducted by the Goldsmith-Nagan Bond and Money Market Letter, now published in the investment newsletter Reporting on Governments.
Froot investigated how future interest rates change in relation to changes in the forward rate. The forward rate is the yield to maturity from a Citron strategy involving zero coupon bonds. If the expectations hypothesis were valid, then interest rates should move one-for-one relative to changes in the forward rate. Froot decomposed the relationship into three components.
• The effect of prediction errors by investors
• The effect of changing term premia, that is, the compensation required by investors to compensate them for the risk of holding bonds
• The one-for-one effect predicted by the expectations hypothesis
Froot confirms that the expectations hypothesis indeed fails. Yields move significantly less in response to forward rates than they would in a one-for-one relationship. Moreover, Froot finds evidence that biased forecasts and varying term premia are both involved. He concludes that for short-term securities, the spread’s bias may be entirely due to term premia. However, he also concludes that the “inability of the spread to forecast future long-rate changes is attributable entirely to systematic expectational errors. … [E]xpected future rates underreact to changes in the short-rate” (p. 304).
Money Market Services (MMS) is one of the major firms that track interest rate forecasts. MMS surveys investors and economists weekly to obtain forecasts for three maturities and two time horizons. The maturities are three months, two years, and thirty years. The horizons are one week and four weeks. Using weekly data from 1988 through 1997, I find that the forecasts obtained by MMS display the same characteristics reported by Froot.20
The MMS interest rate forecasts display a number of interesting features. Suppose we look at the implied forecasts of interest rate changes over the ten-year period 1988–1997. One striking pattern is that for the three-month and thirty-year maturities, forecasted changes were too high relative to actuals. The consensus expectation was that interest rates would be higher than turned out to be the case. For example, forecasters predicted that on average the three-month Treasury bill rate would go up by 13.7 basis points every four weeks, whereas it actually declined by 0.46 basis points. But for the two-year maturity, the reverse was true: the two-year yield fell by less than was expected.
This pattern did not hold in 1994, when interest rates were higher than predicted no matter what the maturity or time horizon. For example, forecasters predicted that on average the three-month Treasury bill rate would go up by 14.5 basis points every four weeks, whereas it actually rose by 19.4 basis points.
And how about expectations about long rates, whose sharp rise throughout 1994 led to the Orange County bankruptcy? Remember that in December 1993 Merrill Lynch fixed income strategist Martin Mauro had predicted that long yields would fall in 1994. Well, Mauro was not alone. This is also what the MMS forecasters predicted for both two-year and thirty-year yields.
Inflation Expectations and the Yield Curve
How important is inflation as a factor in determining yields?21 In chapter 3, I reported on a study by Eldan Shafir, Peter Diamond, and Amos Tversky (1997), which found that although people understand the difference between real and nominal changes, framing issues in nominal terms is more natural for them. Therefore, real changes tend to be less salient.
As far as the yield curve is concerned, appreciating the role of inflation is crucial. In a most insightful article, Werner De Bondt and Mary Bange (1992) discuss how the yield curve is affected by expectations about inflation. Their study is based on inflation forecasts for the period 1953 through 1987 that were collected by Joseph Livingston of the Philadelphia Inquirer. Beginning in 1946, Joseph Livingston conducted a biannual survey in which he asked respondents to estimate the future rate of inflation over both the next six months and the next twelve months. I have updated the De Bondt–Bange data through June 1998.
A critical question for the determination of yields concerns expectations about inflation. If Investors expect high future inflation, they will bid yields up. Hence, a major issue involves the question of accuracy. How accurate have inflationary expectations been?
Figure 14-4 portrays the time series of the six-month-ahead consumer price index (CPI) forecast error, expressed as a percentage of the CPI. The predictable component of an efficient forecast should, of course, be zero. But De Bondt and Bange find that the forecast error in the Livingston survey has a very strong predictable component. It tended to be negative during the 1950s and 1960s, indicating that forecasters underestimated future inflation during this time. The error turned positive in the early 1970s, at the time President Nixon imposed wage and price controls. It turned negative again in the late 1970s, and then entered a phase where it has been positive most of the time. Since 1980, forecasters have consistently overestimated the rate at which the CPI would rise.
Notice that in figure 14-4, inflation forecast errors tend to be negative during periods of rising inflation and positive during periods of falling inflation. De Bondt and Bange argue that this is an underreaction phenomenon. That is, investors underreact to the most recent decline in the inflation rate. Instead, they place too much weight on historical rates.
De Bondt and Bange make a convincing case that errors in inflation forecasts are the major reason why the expectations hypothesis fails. In particular, they focus on the effect of inflation forecast errors. During the 1980s, both long-term rates and spreads were high because investors were too pessimistic about inflation. As investors received positive but surprising news about declining inflation, yields on long
Figure 14-4 CPI Forecast Error as Percent of Actual CPI, June 1954–June 1998
The time path of the percentage CPI forecast errors between 1954 and 1998 in the Livingston survey. Efficient forecasts fluctuate randomly about zero. But the Livingston errors stray from zero for long periods. During periods of declining inflation they tend to be positive, while during periods of rising inflation they tend to be negative. The pattern suggests that investors underreact to changing inflation: Their predictions for the CPI are too high when inflation is declining, and too low when it is rising. maturities declined. But because investors continued to underreact, and inflation rates continued to decline, yields on long-term securities tended to fall, not rise, as the expectations hypothesis would predict. Moreover, short-term rates fell over the life of long-bonds issued in the 1980s.
Did investors in the 1990s continue to be surprised by how quickly inflation declined? As can be seen from figure 14-4, the answer is yes. Here are the opening paragraphs from an article that appeared in the Wall Street Journal on September 17, 1998.
Investors sent stock and bond prices soaring in one of Wall Street’s strongest days this decade.
The U.S. Treasury market had its biggest rise this year, with the bellwether 30-year bond soaring 2 5/32 points, or $21.563 for each $1,000 face amount. The yield on the issue, which moves in the opposite direction of the price, fell to 6.40%. That was the second-largest single-day decline in the yield in the 1990s.
The plunge in interest rates unleashed a torrent of buying among the big blue-chip stocks that have been out of favor for the past six weeks. …
Behind the powerful rallies was the August consumer-price report, which erased any fears that inflation was about to spring up. This ignited what some traders called a “buying panic” in both stocks and bonds.22
Think about this excerpt within the context of figure 14-4. Does it seem like investors continued to underreact to the decline in inflation? It’s not that people don’t learn. It’s that they learn very slowly.
Full Circle
We are coming full circle. What about a Citron strategy? Did it make sense? De Bondt and Bange’s findings provide an answer, but it depends on both the magnitude of the spread and past forecast errors about inflation. When spreads are high and investors have been under-reacting to
declining inflation, then a Citron strategy indeed does make sense. This was the case throughout the 1980s and 1990s.
De Bondt and Bange report that there are three important features concerning the link between predictable interest rate movements and inflation forecast errors. First, when the yield spread is above average, inflation forecasts tend to be too high. Second, when the twelve-month-ahead inflation forecast exceeded the six-month-ahead forecast, investors subsequently earned positive excess returns by holding long-term bonds. Third, past inflation errors are positively correlated with future excess returns. This means that not only are excess returns forecastable, they can be predicted by the extent to which investors misjudged recent inflation.
A Citron strategy makes sense because it capitalizes on investor errors. But as we saw very clearly, the strategy is also risky. Why? Not because of uncertainty about what the Federal Reserve might do, but because the strategy’s success relies on particular investor errors. In his 1995 survey article, Campbell emphasizes that the 1994 surprise exceeded the extent of Federal Reserve tightening at the short end of the yield curve. In addition, long yields rose by about the same amount as short yields. What hurt the Orange County Investment Pool was the fact that Citron both guessed wrong and chose to be highly leveraged.
Of course, this begs the question as to why long-term rates rose in 1994. Was it heuristic-driven bias? Both Froot and De Bondt–Bange state that as a general matter, long yields do not overreact to changes at the short end. I suppose you could say that 1994 was the exception that proved the rule.
Beyond Greed and Fear Page 27