Fair value accounting: information or confusion for financial markets?

For specialized firms (pension funds, investment banks, etc.), fair market value applies to all assets, liabilities, or both. Moreover, while SFAS 115 does not apply to unsecuritized loans, it does apply after their securitization.

The rationale for such an option is well described in Citicorp’s financial statements: “The fair value option provides an opportunity to mitigate volatility in reported earnings that resulted prior to its adoption from being required to apply fair value accounting to certain economic hedges while having to measure the assets and liabilities being economically hedged using an accounting method other than fair value.” (Citicorp’s 2009 annual report, p. 138).

The latter point is difficult to ascertain. For instance, Fiechter (2010) finds evidence that banks applying the Fair Value Option (FVO) with the intention of reducing accounting mismatches as well as banks that apply the FVO to financial liabilities report lower levels of earnings volatility than non-appliers. However, such reduced volatility either reflects calm and steady underlying economic conditions or, alternatively, increased ability by executives to smooth reported earnings through strategic use of the various FVA measurement levels.

The FR Y-9C report form is to be filed at least by bank holding companies with total consolidated assets of $500 million or more. Prior to 2006, the report had to be filed by bank holding companies with total consolidated assets of $150 million. Due to the requirement of at least three analysts following, we do not observe a drop in the number of banks around 2006.

We consider active all those analysts who forecast annual earnings at least twice in a year in two but not in the same quarter.

Please, refer to Sect. 3.6 for the details on how to calculate the precision of public (h) and private (s) analysts’ information.

Untabulated results using panel fixed and random effects linear model specifications show that total fair value is statistically positively associated with accuracy and dispersion.

FV is equal to the sum of available-for-sale securities, loan and leases held for sale, trading assets, other financial assets and servicing assets at fair value, trading liabilities, all other financial liabilities and servicing liabilities at fair value, and loan commitments not accounted for as derivatives at fair value.

The approach chosen in this paper is consistent with the approach used by Ball et al. (2012).

Starting the first quarter of 2007, banks have to disclose fair value levels to the Federal Reserve Bank of Chicago.

Results are unchanged if we use the entire sample period.

We hand-collect data on fair value levels and disclosure from annual and quarterly reports, because the Federal Reserve Bank of Chicago required the disclosure of Level 1 assets and liabilities only in 2008. We exclude 19 banks due to unclear disclosure or missing values.

Using the mean earnings per share estimated by analysts does not affect the results.

Standardizing the FV over total book value of equity does not affect the results.

We are aware that, as in most studies that analyze the effect of a regulatory change, it is not possible to eliminate the influence of other changes. In our case, for example, the forms FR Y-9C were 39, 40, 43, and 51 pages long in 2006, 2007, 2008, and 2009, respectively, with additions that include some FV level information in 2007, details on the nature of loans measured at fair value (real estate, commercial, or industrial, individuals) in 2008, and details on the nature of assets and liabilities in 2009 (from three categories for the assets at FV in 2008 to 7 in 2009). Even with this caveat, we are using SFAS 157 as the most prominent change related with fair value accounting around 2007.

In their analysis, Song et al. (2010) include the amount of nonfair value assets. To capture the effect of banks’ investment decision, we include the loans at the amortized cost in our tabulated results. However including the nonfair value assets instead of loans does not change the results.

We adjust Amihud’s (2002) measure of single stock illiquidity to the entire US market: the quarterly average ratio of the daily absolute return to the (dollar) trading volume on that day.

Results are robust if we perform our analyses with the same models but using only one of these two variables. To mitigate the effect of short period change in GDP, we also perform the very same models using quarterly change in GDP with respect to the same quarter one year before (ΔGDP _y ). Although results using both unemployment and ΔGDP _q or ΔGDP _y are the same, this last variable is highly negatively correlated with unemployment (−0.764).

For additional details on the application of fair value accounting in bank holding companies in the United States, see Nissim and Penman (2007).

In untabulated tests, we exclude each year of data one by one, and in all of the tests, the results are the same.

The regression models use a subsample of observations centered around the application of SFAS 157. Therefore we consider the years 2007, 2008, and 2009 as the post-adoption period and the years 2004, 2005, and 2006 as pre-adoption period. Considering the entire sample from 1996 does not modify the results.

Note that SFAS 157 mandates the disclosure of fair value levels starting from the fiscal year beginning after November 15, 2007. The adoption period overlaps quite extensively with the start of the financial crisis, which may induce analysts to screen accurately securities typically evaluated at fair value. To alleviate the concern that the results are driven by an increase in the level of attention paid by analysts, we include in the post-adoption period the bank-quarter observations that voluntarily release fair value. By including banks that voluntarily reported fair value levels in 2007, before the most severe crisis period, we aim to reduce the concern that the results are simply driven by analysts’ attention. We are aware that banks that report on a voluntarily basis have incentives to be more transparent and their commitment towards transparency may drive the results. In any case, using only 2008 and 2009 as the post-adoption period does not affect the results. Finally, results in Table 7 (columns 1–4) show that, when the illiquidity and the VIX are high, the accuracy of analysts’ forecasts does not increase, thus reducing the concern that the increase in accuracy during the post-disclosure period is due to an increase in the level of attention paid by analysts.

Altamuro and Zhang (2013) find that Mortgage Servicing Rights (MSR) measured at level 3 are more informative than MSR measured at level 2. However, most banks do not report any MSR, and these represent a minute portion of banks’ assets (less than 1 %), thus making any generalization to other classes of asset tentative (Hendricks and Shakespeare 2013). In contrast, consistent with our results, Evans et al. (2014) show that the predictive ability of FVA-based information on interest-bearing investment securities is positively related to its measurement precision, with level 3 exhibiting the least precision (vs. levels 2 and 1). Our results provide also further support to Bratten et al. (2012), who find that the impact of FVA on the predictability of future return on assets, cash flow from operations, and earnings is conditional on several variables or mixed at best.

Our results can be interpreted as the effect of a low level of reliability of the disclosure of fair value of loan portfolios. However, Nissim (2003) suggests that banks are prone to overstate reported fair value of loans with the intended goal of masking risk and performance. Our results show that negative differences (fair value lower than the amortized cost) worsens the information environment, which is contrary to what Nissim (2003) would predict.

Focusing on loan portfolios, Cantrell et al. (2011) provide evidence that is consistent with our findings as they show that historical cost information is better predicts future credit losses, both short and long term, and bank failures than FVA.

For the purpose of computing the required level of regulatory capital, all asset categories are assigned a risk weighting that varies between 0 and 100 %. Assets with a risk weight of 0 % (typically US government bonds) do not require a bank to hold any regulatory capital: in essence, a bank can hold as much of this type of asset as it wants without putting up capital to support that investment. In contrast, assets with a risk weight of 100 % must be backed up with the required level of regulatory capital. For example, if the required capital ratio is 10 %, then $100 of assets with a 100 % risk weight require the bank to hold $10 of capital (10 % × $100 × 100 %).

Untabulated analyses reveal that the results are qualitatively the same if Illiquidity and VIX are dichotomized using the median.