Dr. Ruggero Scorcioni’s recent article on FactSet’s High Impact Transcripts (HIT) signal, offers an exciting application of machine learning to active equity portfolio management. As Dr. Scorcioni noted in his article, HIT is a signal that classifies a company’s earnings transcript to its likelihood to outperform (+1) or underperform (-1) the long-term trend of a company’s stock price. This article is a small excursion into the application of HIT for fixed income as well.
It has long been acknowledged that equity returns have an impact on the associated corporate bonds. However, measuring this impact is tricky. First, bond prices are subject to a variety of influences on returns such as the passage of time and changes in interest rates. Thus, the focus should be on returns due to spread changes. However, the impact of equity returns on corporate spreads is still not as direct as desired. Corporate bond spreads also reflect a bond’s relative liquidity. Taken all together, the best we can hope for across a bond index is a causal indication. The following research derives such an indication, although more work is required to confirm its consistency and magnitude.
The search itself was relatively simple: match HIT scores over 2015-2019 for the bonds issued by the constituents of the S&P 500 index. The starting point is the average spread across all bonds of the issuer for each HIT incident. The test statistic is the ratio of the future 20-day average spread versus the prior 20-day average spread, assigned to HIT scores of +1, 0, and -1. Like most things in fixed income, we would expect the impact to be backward. That is, companies with stronger earnings will experience lower spreads due to their increased likelihood of survival, while companies with weaker earnings will experience higher spreads as compensation for their increased likelihood of default.
Because spreads are typically quoted in basis points, the table below is designed to highlight the differences across scores. In addition, because these are bonds, the magnitude of response to a +1/-1 HIT score varies by bond and issuer; therefore, we need to normalize the +1/-1 HIT scores to their change versus the HIT=0 score. Specifically, the +1/-1 HIT scores shown are the percentage change versus the HIT=0 score. The table and graph below show the results AA-A, BBB, and BB-B rated bonds.
|
|
HIT=1 |
HIT=-1 |
|
Rated AA-A |
-1.6 |
0.1 |
|
Rated BBB |
-2.0 |
1.4 |
|
Rated BB-B |
1.4 |
0.6 |
Source: FactSet
What is interesting here is the difference between investment-grade bonds and high-yield bonds. High-yield bonds have a clear negative slope, indicating that positive company earnings imply greater odds of survival, and thus tighter spreads. However, for investment-grade bonds, we see positive company earnings indicating wider corporate spreads.
If we check the frequency of HIT scores, we see that the average company over this time frame has approximately 18 HIT scores. However, 31% of the HIT scores for BB-B securities were either +1 or -1, compared to 10% and 3% for BBB and AA-A bonds, respectively. The implication is that the greater leverage of high-yield bonds makes their earnings, and the reaction to their earnings announcements, that much more volatile.
In summary, initial indications indicate that the HIT score is also useful for corporate bonds as well as equities and is significant as a leading indicator for the frequent spread movements of high-yield bonds. Further research is required into the impact on investment-grade bonds and the circumstances where the HIT score has the greatest impact. However, the potential is exciting.