One of the industry trends that is impacting both asset managers and asset owners is that several vendors are beginning to offer “all in one” centralized computer systems that are intended to support all functions of an investment institution including decision making (“front office”), trading (“middle office”) and operations (“back office”). The argument for these all in one systems is that there are material operating cost savings associated with consolidating many independent systems. On the other hand, several clients have expressed the view that while these systems may offer “good enough” functionality in areas such as risk assessment or transaction cost assessment, they are uncomfortable making decisions on what they believe is less than the best available analysis.
It can be clearly demonstrated that in most cases, any reduction in the perceived accuracy of risk assessment for a large asset owner will vastly outweigh the economic benefit of the cost savings. A simple example should be sufficient to illustrate the point. Let us consider the case of a hypothetical asset owner called Pension X, which has $30 Billion in assets of which 50% are invested in illiquid assets (which are problematic for most risk systems). Pension X currently spends $120,000 per annum for a risk system but believes they can save $25 million per annum in operating costs by consolidating on an “all in one” system. We will use very typical capital market expectations that the expected rate of return on their portfolio is 7% (arithmetic average) with an annual expected volatility of 10%.
Assuming a fixed level of expenses, the expectation of the percentage geometric mean rate of return is 6.5% (7 – 102/200) [see Messmore (1995)]. The current spending of on risk assessment in dollars is 1.2 * 105 which as a percentage of the fund is 4 * 10-4. If these costs were removed, the expectation of the geometric mean return would increase to 6.5004%. This doesn’t seem very significant but saving $120,000 per annum certainly sounds appealing as compared to not doing so.
Now we must consider the economic cost of changing risk assessment systems. Prior to this consolidation effort, Pension X had chosen to use a particular risk assessment system presumably because they believed it was the best for their needs. The question at hand is how much accuracy could be lost in the risk assessment process before the cost savings are offset by less effective portfolio management. We can represent this problem by simply adding an “uncertainty” increment into the risk. If we know less about the risks we’re taking, this is equivalent to saying we are taking more risk as we are uncertain as to our situation. This question is easily solved. Our improved return of 6.5004 now must now be greater than (7 + .0004 – (10 + U)2 / 200). Solving this equation for U at .0004, we realize that the benefit of saving $120,000 per annum is offset by carrying combined volatility and uncertainty of 10.0004 or greater, which is just .04% of the 10% initial volatility. Even if we assume that the uncertainty was concentrated in the half of the Pension X portfolio that was illiquid, an uncertainty increment of just .08% is equally damaging. Reducing the accuracy of the risk system by even by this tiny margin outweighs the cost saving for Pension X.
Let’s consider an even more radical idea. How much would the risk assessment accuracy have to decline in order to offset the entire $25 million in annual savings that are expected by consolidating systems? Repeating the same algebra, we see that $25 million is 2.5 * 107, against a portfolio value of 3 * 1010, the fraction is 8.3 * 10-4, or .083%. Solving for U once again, we get .79, if the incremental uncertainty associated with switching risk assessment method is 7.9% or greater of original assessment of 10%, the entire $25 million in annual cost savings in nullified.
However, there is another thing we’ve left out of the analysis. Everything up to this point assumes that Pension X is a “growth optimal” investor with an infinite time horizon. In Northfield terminology it assumes a “Risk Acceptance Parameter” or RAP of 200. If our assumed volatility of the Pension X portfolio of 10% is about right, this would imply a RAP value closer to 60. This value is derived from the “discretionary wealth hypothesis” [see Wilcox (2003)]. If we include this consideration in our analysis (Pension X is more sensitive to uncertainty than a growth optimal investor), the incremental risks associated with a less accurate risk system are magnified. With an estimated risk tolerance of 60, the threshold inaccuracy level needed to nullify the entire $25 million savings (i.e. the critical value of U) is .25 or 2.5% of the original volatility of 10. If we double that to assume that only illiquid assets are impacted, the new value is 5%. So, with a realistic assessment of the risk tolerance of Pension X, a 5% decrease in the accuracy of risk estimation on illiquid assets is sufficient to nullify the entire $25 million in annual savings.
The lesson should be clear. For large asset owners, the cost of having as good a risk assessment as they can get pays for itself hundreds of times over. Northfield has made this argument many, many times over the years. For example, http://www.northinfo.com/documents/204.pdf from the Newport seminar in 2006 (see slide 10). From a business management perspective this sort of behavior is a classic case of trying to control what you can fully control (i.e. expenses) without understanding the impact on what you can’t fully control (risk). It’s like the old joke where a man loses his car keys on one side of a parking lot but walks across and begins to look for the keys on the other side of the parking lot. When this odd behavior is questioned by passers-by, he replies “the light is better over here.”
A downloadable version of this essay is available on Northfield’s website:
