Ahead of Saturday's Berkshire Hathaway (BRK.A) (BRK.B) annual meeting, we're taking a closer look at the best way to value the complex company. We believe that understanding the benefits and shortfalls of different methodologies can provide valuable insight into the ways in which different investors are approaching the firm's overall valuation. Part 1 of the series, on an earnings-based multiple approach, is available here. Part 2 on book value can be found here. Part 3 on the two-column approach is here.
The float-based approach to valuing Berkshire is fraught with its own issues
The approach we will examine here for discounting the value of the investment portfolio is the float method. Over the years, the terminology surrounding this approach has become mixed to the point where many people use the terms float-based as a synonym for the two-column approach. The method we will investigate is similar to the one originally outlined by Alice Schroeder at PaineWebber in the late 1990s. This approach seeks to capture the premium of the insurance business over book value by capitalizing the future cash flows from the insurance float (the excess of premiums paid by policyholders that have not yet been paid out as claims).
We have calculated a spread, representing an estimation of Berkshire's long-term returns on its float. In this case, the cost is negative, indicating an expectation of an underwriting profit. The hypothetical yield on the current value of the float is then capitalized in order to generate the estimated premium over book value. As this approach theorizes that the value of the float represents a premium over book value, adding the statutory capital assigned to the insurance business less the carrying cost of Burlington Northern Santa Fe yields the final result. As originally suggested by valuation work done by Ravi Nagarajan at Rational Walk, we subtracted the purchase price for BNSF as its shares are held at cost on National Indemnity's books. Including this in the calculation would lead to double counting.
This insurance and investment value is then combined with a multiples-based approach that estimates the worth of the noninsurance businesses. Although we believe that hypothetically this is an improvement versus taking investments at their full value, there are a number of problems. First, the model is very sensitive to small changes in assumptions. Since we are essentially assuming an immediate perpetuity by capitalizing the hypothetical earnings from the float, the approach requires overly conservative assumptions. Also, assigning a precise discount rate is problematic. Again, since small changes have a significant impact on our conclusions the uncertainty is meaningful. Finally, and most important, using this framework does not seem to give appropriate values to peer insurers, as shown in the table below.
This approach results in estimated intrinsic values similar to the approach that takes investments at 100% of their carrying value. Besides the problems we noted above with applying the float-based approach to Berkshire, we notice certain debatable assumptions that were used in the original implementation of the float method that can lead to erroneous results for insurance companies. Most important, we question whether using the risk-free rate as a discount rate is appropriate. By definition the discount rate should be that return that investors require on a project, and we believe it is unrealistic to assume that investors would only require the risk-free rate given the volatility of both the insurance and investment earnings.
Instead, we propose using our estimate of the cost of equity that reflects these risks. Using the risk-free rate as a discount factor is aggressive and may have led other practitioners to use overly conservative estimates on other model inputs. We believe it is important to be able to differentiate between insurance companies through the underwriting profitability (shown as the cost of float in this method), which, along with the discount rate, we adjusted.
This approach moves the values in the right direction for these insurers, but in nearly all cases they are still far from either the market price or our estimate of their fair value. In our opinion, there are a number of possible factors that may lead to this conclusion, most of which are due to the simplifying assumptions necessary to complete the model combined with its lack of flexibility. Since we are calculating the value of each insurance company through an immediate perpetuity we lose the ability to forecast interim periods of excess or subpar profitability and growth. Also, while it attempts to mimic the cash flows of the insurance operations, the model does not account for excess or deficient capital levels or leverage, which may be an important valuation factor for assessing the firm's value to shareholders.
As expected, we got results that more closely align with our fair value estimate from either discounting or using the excess investments above the firm's float instead of taking the full value of per share investments. The float method seems to be improved by our alternate approach, but, as mentioned earlier, there appears to be flaws with its applicability to other insurers. This indicates that its proximity to our fair value estimate may be due to coincidence or unique circumstances at Berkshire rather than from the approach's fundamental accuracy. For all of the methods, one must still decide on an appropriate multiple to earnings for either the rest of the business or the rest of the business, excluding insurance operations.