Transcript of Larry Summers NBER Macro Annual 2018 I salute the authors endeavor to use market price to examine the riskiness of the financial system and to evaluate the change in the subsidy represented by government guarantees. As illustrated by my work with Natasha Sarin (2016) 1 that the authors reference, I believe that market information is at a minimum a valuable complement to accounting information in evaluating the health of banks. I would guess that their broad conclusions that if a crisis like 2008 were to happen again, we would have insolvent banks is correct. And I find it plausible that as the authors believe a combination of more regulatory capital, establishment of resolution procedures and official commitments to move beyond too-big-to-fail have reduced the market s perception of implicit guarantees. That said, I have to report that I m almost entirely unconvinced by any of the author s estimates and believe that all reflect arbitrary and in some cases implausible modeling assumptions. I do not believe they have any real basis for their claims about the extent of franchise value as opposed to capitalized government subsidies. Its not that I have clearly different views than the authors, just that I do not believe their measurements are convincing. First, there are some real questions about the theory of subsidies that are raised by the kind of analysis that is done here. Let s imagine that the government decided to subsidize ice cream for all companies that sold ice cream cones. What would we expect? I think we would expect that there would be lower priced ice cream cones. I think we would expect that the quantity of ice cream cones sold would go up. I think we would expect no change in the Q-ratio of ice cream cone companies, if this was a competitive industry. If ice cream cones, and the production of ice cream cones, involved investment that took place with adjustment costs, we would expect some brief interval of higher profits before the industry expanded, but to the first approximation, the change in the Q-ratio would be very small compared to the magnitude of the change in government subsidies. Of course matters would be different if ice cream cones were produced by a monopolist or if there were barriers to entry into ice cream cone production or limits on the expansion of ice cream cone producers. Then one would expect more capitalization of the subsidy into the market value of ice cream companies. Suppose an economist who for some reason didn t know the size of the subsidy sought to infer it from ice cream cone company stock prices. I do not think she would be taken seriously. There would be little basis for knowing how much of the subsidy was capitalized and how much in the market s view would be competed away. And there would be many other things happening economy-wide that would affect ice cream cone company stock prices. Now, let s suppose that the key input into the production of bank loans is bank liabilities, and let s suppose that the government subsidizes bank liabilities. Then, if one assumes that the banking industry is competitive, it seems to me, one would reach an essentially parallel 1 Sarin, Natasha, and Lawrence H. Summers. "Understanding bank risk through market measures." Brookings Papers on Economic Activity 2016.2 (2016): 57-127. 1
conclusion, that expanded subsidies from the government would lead to competition between all the banks, which would lead to expansion by the banks, but it would not lead to an increase in the long term profitability on assets of banks. Now, if banks were small, compared to the loan industry, and, so, the pricing of loans didn t change, things would be different. Similarly, things would be different if the price of ice cream cones was fixed and only one small sector of ice cream cone manufacturers were subsidized. But that would surprise me as the right assumption with respect to bank loans. If the whole banking industry were a monopoly, one would expect that my analysis was wrong, as with ice cream cone monopolists. I would guess that banks are closer to competitive than to being long term monopolies but the key point is that, as with ice cream cones, inferring the value of the subsidy from stock prices is implausible. In Appendix Section B5 the authors attempt to respond to this issue, but I have trouble following their argument. Second, assume away this set of problems. It seems to me, there are a variety of more specific reasons to doubt the estimates the authors provide. First, when Sarin and I were doing our work, I thought a lot about changes in the too-big-tofail subsidy and using option theory to deal with this aspect. I decided that if the government operated the way it would in the model, which is the government properly watched banks closely, watched the assets, watched the liabilities, was in a position to liquidate the assets and liquidate the liabilities, and the government liquidated the bank as soon as the assets equaled the liabilities; then except for jumps, there would be no cost to the government guarantee, and, so, it was all about jumps and there really weren t that many jumps, in practice. After all, stock prices rarely move 5 percent in a day. An actual valuation of the liability guarantee has to recognize that the government is not that smart, that the government wasn t that good at measuring these things precisely, that the government was heavily influenced by politics, and, therefore, that the government would tend to be late to the party in recognizing asset value declines, particularly on illiquid assets like bank loans. There is also the point that the government would not be able to realize the same value from the assets that the banks would as an ongoing concern. Why should you repay an institution that s about to close and will never be in a position to lend you money or do anything to you again? So, the value of the put option represented by a government liability guarantee depends crucially on its horizon, which in turn depends upon these heavily institutional questions; how slow the government is, and what are the problems with the liquid markets assumption? It is right to apply option theory in principle. But I can t imagine why one would suppose that a two-state model that assumed that one was in a crisis or one was not in a crisis, that capital was checked annually and that at some moment like the trough of the 2008 crisis liquidation was possible were reasonable approximations. Option theory when you do not know the strike price of the option or its horizon is not actually very helpful. 2
Second, I will confidently predict that if you just looked at the stock market, in real terms, it would track fairly accurately Tobin s Q for the whole economy and the ratio of the market-value to the book-value of banks. This suggests that much of what is driving fluctuations in bank market-to-book ratios is broad economic factors. The authors suggest otherwise. This is because they do not reckon with the differences between marking to market and book value accounting. I would be very surprised if the fair value calculation that s used in the authors calculations corresponds to an actual market value of loans when loan sales were attempted in 2008. The authors procedure in effect, smooths franchise value very substantially and attributes the rest of the fluctuation in market value to fluctuating values in the government guarantee, and, then, not very surprisingly, concludes that the fluctuating values of the government guarantee are a large part of what is going on. Third, I was also struck that a substantial part of the risk in the banking system must have to do with duration mismatches. Certainly, if one looks at specific institutions, there must be a number that at any given moment have substantial duration mismatches. This further reduced my confidence in the calibration. Last, I would simply suggest that it is almost impossible for me to believe that there was not a substantial behavioral element in the valuation of bank stocks in the run-up to the 2007 period. So, to somehow suggest that all of the valuation is best thought of as either a rational calculation of franchise value or as a value of a government guarantee seems to me pretty substantially unlikely. Sarin and I recognized that it was a basically uninteresting comparison to look at various measures of bank volatility, or various measures of required returns on preferred stock in the current period and compare those with the 2005 to 2007 period, which we took to obviously be distorted by investor misperceptions. We thought the interesting comparisons were with the pre- 2005 period that were plausibly without those misperceptions or at least less of them. For all these reasons, the authors effort to calibrate the importance of franchise value versus government subsidy value is unconvincing, even if I waive aside the fact that I would expect the principal effect of the subsidies to flow into lending rates rather than bank stock prices. 2 Additionally, there are some subtle issues that I do not feel I fully understand having to do with the relationship between franchise value and bank capital. The authors write and at times Sarin and I wrote as if franchise value was a perfect substitute for bank capital. If the bank has more capital, it s a sounder bank. If the bank has more franchise value, it s got more capital, so it s a sounder bank. However, if a bank goes out of business it likely loses its franchise value. And, so, if I am thinking about my risk in lending to a bank then if the bank is sitting with assets that it can sell, that is reassuring to me about the health of the bank. If the bank is sitting with franchise value, whose existence depends on people like me all not staging a run, that is at least a more problematic scenario. And, so, the question of what the relationship is between franchise value 2 The authors attempt to respond to some of these points in Appendix B. But I do not think they provide convincing defenses of their option horizon assumption, their use of fair value accounting, or their ignoring behavioral aspects of bank stock pricing. 3
and economic capital and its relationship to stability is delicate. Franchise value is an illiquid asset. I don t think the treatment in this paper is quite right. It s clear to me that it s better to have franchise value than it is to not have franchise value, but it sure feels to me like it s better to have a dollar of book value of equity than to have a dollar of franchise value. Just how one compares these, I m not sure. There s probably a big distinction, here, between big banks and small banks. If you re a little community bank and your franchise value takes the form of a bunch of sticky deposits, and if you fail, you re going to get sold to a somewhat bigger bank that s going to continue to have those deposits, so your franchise value is fine. If you re a large institution that cannot be sold, whatever your franchise value exactly is, it seems to me that it is a much more problematic thing from the point of view of your stability. The last thing I would say is, and I think the large question that I don t think this paper resolves, and that I certainly don t think that Sarin and I resolve, is how best to integrate market value information and book information in assuring bank stability. I am reasonably sure that relying only on the market is foolish. You would conclude on the basis of a variety of measures pricing of subordinated debt, option value calculations based on the volatility of equity for example that the banking system was extremely safe in 2007. So, I m fairly sure that that is a serious mistake. I am almost equally confident that ignoring market value information and relying on book accounting value is an egregious error. I would note that the chairman of the SEC proudly proclaimed that Bear Stearns capital ratio was 18 percent one week after it had failed. I would note that the relevant authorities professed Wachovia, WaMu, and Citigroup all to be wellcapitalized and without problem as measured by capital ratios on the brink of their failure. I would note that surely the most egregious failure associated with the financial crisis, and one of the least remarked, is the failure of the authorities to do anything to stop dividends or force capital raising in the period from spring 2008 to fall 2008, when much could have been done and when markets were sending fairly clear signals that all was not well. So, it seems to me it is clear that some kind of two-key system is necessary, where regulators become alarmed either if accounting information provides a ground for alarm or if markets provide a ground for alarm, but as to how that best can be implemented, I have not seen what I would regard as substantially practical proposals. And I would note, finally, that anyone who believes that this problem is in-hand should ponder the fact that our regulatory authorities, as a group, proclaim each year some form of the proposition that I would regard as being at the edge of absurd. 3 That if the Dow were to fall to 10,000, if house prices were to fall 50 percent more than they fell during the 2008 crisis, if the 3 In fact, Sarin and I are working on a follow-on paper that makes exactly this point. A naïve market-based stress test produces outcomes much less sanguine than the Federal Reserve s annual exercise. Sarin, Natasha, and Lawrence H. Summers. "On Market-Based Approaches to the Valuation of Capital." Handbook on Financial Stress- Testing, forthcoming. 4
unemployment rate were to reach double digits, that no major financial institution would suffer a loss of more than five percent of its capital or be in any need of raising capital. 4 Whether we need to have a financial system that is fully insured against that contingency or whether we can rely on some broader public insurance for that fairly rare contingency is, I think, a legitimate question for debate. But to suppose as the Federal Reserve stress tests do that all major institutions could weather a storm much worse than 2008 without needing to raise capital strikes seems to me highly dubious. Certainly extrapolations from market information on banks health suggests cause for grave concern. All is not in hand. 4 See Dodd-Frank Act Stress Test 2018: Supervisory Stress Test Methodology and Results June 2018 for a description of the severely adverse stress scenario. https://www.federalreserve.gov/publications/files/2018-dfastmethodology-results-20180621.pdf 5