Title: Navigating Bid-Offer Valuation Adjustments Amid Financial Market Turmoil
In March 2023, the financial markets were rocked by the rescue of Credit Suisse, the collapse of Silicon Valley Bank and Signature Bank, the shutdown of crypto lender Silvergate Capital, and extraordinary moves in global fixed income. These events serve as a stark reminder of the market's capacity for shock and awe. For control functions looking after derivatives portfolios, these events also created pressing questions for Q1 2023 quarter-end processes, adding to the already complex tasks associated with routine quarter-ends. One crucial question was whether the bid-offer valuation adjustment (bid-offer VA) on the portfolios should be increased to account for the heightened challenge of hedging risk in the current market environment. In today’s post, we will examine market standard procedures for calculating bid-offer VA, the challenges associated with implementing these policies, and potential future improvements.
The bid-offer VA is an essential calculation for many financial institutions, as it ensures that a portfolio's valuation reflects the cost of closing out the residual risks of that portfolio. The valuation control process typically begins by estimating the mid-market value of each trade in the portfolio and using their sum as an approximation of the total portfolio value. Since this initial estimate relies on mid-market values, the valuation control process must include a subsequent computation to factor in the cost of hedging risks – this is where the bid-offer VA comes into play. The exact methodology for calculating this adjustment varies among financial institutions due to differences in measuring and categorizing risks. However, a useful way to conceptualize the calculation is as the sum of the costs for closing out every individual risk factor in the portfolio. The close-out cost for each risk factor is determined by the magnitude of the risk multiplied by half of the bid-offer spread for that specific risk.
Market observations determine the bid-offer spread for a risk factor, and it is this unpredictable element that has raised concerns this quarter-end. Bid-offer spreads tend to spike during periods of market sentiment driven by contagion fears, as market-makers increase the buffer on their quotes to avoid being left with an unhedged position when the next market-moving event occurs. This phenomenon can be observed in the U.S. interest-rate swap (IRS) curve, which experienced significant turmoil in March 2023 as market participants rapidly re-evaluated the path of Federal Reserve rate decisions, decreasing the expected terminal rate by 100bps within just a few days.
We analyzed publicly available IRS transaction data to compute the implied bid-offer spread for 1y, 2y, 5y, 10y, and 30y tenor SOFR swaps with standard market conventions. Figure 1 illustrates how the implied bid-offer spread for these instruments has changed since December 2022. It highlights the market's intense apprehension, with the implied bid-offer for 1y and 2y swaps skyrocketing quarter-on-quarter by a multiple of 6 and 3, respectively. The market's focus on central banks' near-term responses resulted in a more subdued reaction for longer tenor swaps.
The economic implications for interest-rate derivatives portfolios are substantial. A hypothetical short-end IRS trading book containing only 1y and 2y tenor swaps that did not put on new risk between year-end 2022 and Q1 2023 would see its bid-offer VA increase by 3 to 6 times. Such a shift in VA typically sparks two opposing arguments. Some may contend that the events involving Credit Suisse and others were idiosyncratic and did not fundamentally alter the IRS market, so increasing the bid-offer VA based on this spike in spreads would not accurately represent the cost of closing out the interest-rate risk in the portfolio. A highly volatile bid-offer VA can dominate the economics of the client franchise in the short term, e.g., trading books might turn away otherwise profitable client business if spikes in bid-offer VA overwhelm the value of the client trades. Conversely, it might be unwise to entirely disregard the market's current signals. In fact, increasing the bid-offer VA could be a prudent move to encourage a reduction in overall market risk in case the contagion worsens.
Experienced valuation control groups have dealt with these types of internal discussions multiple times in the past as extreme market shocks have become more common place since the Global Financial Crisis, e.g., during the European sovereign bond crisis (2012), the taper tantrum (2013) and, of course, the Covid-19 pandemic (2020). In our conversations with these teams, most agree that a common-sense approach would be to make the bid-offer VA reflect current market signals but also respect the fact that bid-offer spreads should be more stable than other market data points such as the actual underlying prices. A natural way to do this is to update the bid-offer spreads as frequently as possible and do some sort of averaging over the most recent observations.
To simulate this, we examined four different policies for updating the bid-offer spread: quarterly (Q), quarterly with averaging over the last quarter and the current quarter (QA), monthly (M), and monthly with three-month averaging (MA). We used the example of a book with a $50k pv01 exposure in 1y tenor and $50k pv01 exposure in 2y tenor.
We then generated Bid-Offer VA for this example book considering three scenarios taking place over the next three months:
1. Bid-offer spreads persist at March 2023 levels.
2. Bid-offer spreads quickly revert to year-end 2022 levels.
3. Bid-offer spreads gradually decrease to a level between year-end 2022 and March 2023 over the next three months.
The results are shown in Figure 2. As anticipated, the results reveal that policy Q struggles to keep pace with market spreads when a reversal occurs (i.e., Scenarios 2 and 3). Policy M does better on Scenario 3 as the bid-offer VA smoothly follows the decrease in market spreads. However, for Scenario 2, this policy also displays a very sharp up and down which is not the ideal behaviour as we mentioned above. Policy MA seems to navigate all three scenarios well; however, it should be noted that even though the VA under this methodology ends up at the same level as the other methods if Scenario 1 occurs, it does so with about a 2-month delay. This delay could be calibrated, e.g., using a weighted average where more weight is put on the current market spreads will shorten this delay. Finally, policy QA seems to combine the worst characteristics of all the others, i.e., showing significant VA spikes but also being slow to keep pace with the market.
Given that monthly updating with averaging performs better in all three scenarios, it should be a good choice for most financial institutions. However, this is not what gets implemented in practice for most portfolios because updating the market bid-offer spreads is labour-intensive and usually produces unreliable results. For most derivatives, there are traditionally a couple of different ways of observing bid-offer spreads from the market. First, the financial institution could reach out to inter-dealer brokers who would try to find bid-side and offer-side quotes for the product from market-makers. This is a manual process which is hit-and-miss in terms of data quality, e.g., the inter-dealer brokers might not be able to find any market-makers willing to make live prices so the resultant data might be some combination of indicative and stale quotes. Worse still, it might not be clear which of the bid-offer spreads came from tradeable two-way prices versus ones that came from indicative quotes. The second way of updating bid-offer spreads is to use third-party service providers whose primary business is polling financial institutions for data and would provide colour on the average bid-offer spread that the sample set is using based on the polled results. The danger in relying purely on this method is that the financial institutions participating in the polls will have a tendency to make submissions that conform to the mean of the sample set, hence the results could be slow to reflect current market trends.
A third approach to observing bid-offer spreads has recently emerged and uses ideas from machine-learning applied to the large data sets of derivatives transactions that have become more widely available in the last few years. In fact, this is how we computed the transaction-implied bid-offer spreads used in all of the analysis in this post. This approach is showing promise as a new way of updating bid-offer spreads that is both more efficient than speaking to inter-dealer brokers as well as being more reflective of the market than using third-party service provider polls. Specifically, it can be automated as part of a data pipeline so eliminates the manual intervention associated with inter-dealer broker chats or emails. And, because the bid-offer spreads are computed using real transaction data, there is none of the bias that can be introduced in a poll due to the feedback loop as financial institutions adjust their behaviour to the results of the poll. Furthermore, the computations are transparent and easily replicable.
In conclusion, Q1 2023 reminded us once again that the financial markets are experts at coming up with surprises, and these have serious knock-on effects into many aspects of looking after derivatives portfolios. This severely tests many of the control procedures at financial institutions and places a heavy burden on teams like valuation control who already have many time and cost constraints. We believe that innovations like Big Data and machine-learning will provide some welcome relief and we will continue to update on promising use-cases.