Stress Testing ProcessMay 23, 2014 - 16:51
Cube Research offers clear and fair quantitative research on retail structured products.
Our primary aim is to give advisers and investors an indication of the risk and return that each product offers. Our analysis allows products to be filtered to identify a short list of products that may be suitable, and then ranked using a number of different measures to help identify the product that best meets the investor’s requirements.
- We measure risk and return using data that we can observe and which is publicly available through external sources such as Bloomberg
- Our analysis is fully transparent, objective and verifiable
- We are not required to make any assumptions, estimates or judgments
- We use straightforward methodologies, applied consistently across all products
- We offer clear analysis of the risks and returns a product offers
- We calculate the “volatility” of a product and use this volatility to calculate the Risk Score
- We calculate the overall expected return as a percentage per annum that a product offers
- Our analysis makes product comparison straightforward;
- Investors can compare structured products to benchmark equity / bond portfolios
- Investors are easily able to compare structured products from one issuer with another
- Investors can compare products that offer exposure to different underlying markets
- There is no need for specialist statistical or mathematical knowledge to understand the analysis
- Our analysis is consistent with the risk information offered by funds in their Key Investor Information Documents (KIIDs) and consistent with the main financial planning tools and processes
For each product, we offer online analysis and create an Cube Information Document (CID) as a PDF that can be downloaded from our web site. This document is created when the product is first available to investors.
Both the online analysis and CID are updated on a regular basis throughout the term of the product. Up to date analysis will reflect the current product price and the latest levels of the underlying assets.
We test products under 3 different scenarios
- Historic Back-test; an analysis of the risks and returns offered by a product that uses the actual past performance of the underlying assets
- Market Stress-test; the same analysis but on multiple re-sampled paths of the historical returns of the underlying data
- Market and Issuer Stress-test; an adjusted Future Stress-test where the risks and returns are adjusted to reflect the possibility of the issuer defaulting
Historic Back-test methodology
We use daily closing levels of a product’s underlying(s) as far back as the beginning of 1993 to analyse the returns that a product would have offered had it been available in the past.
To illustrate, we describe how on 26th July 2013 we back-tested a FTSE100 6 year Autocall with annual kick-outs from Year 2 and a 50% American soft protection barrier; We use the iShares Core FTSE 100 UCITS ETF (which tracks the FTSE100 index), its price data is available on Bloomberg from 3rd January 1984:
- The first 6-year cycle we look at is therefore 3rd January 1984 to 3rd January 1990
- The second is 4th January 1984 to 4th January 1990 and so on, each cycle starting and ending on a trading day
- The last cycle is 26th July 2007 to 26th July 2013 as this is the last complete 6 year term
- There are 5971 such cycles
For each cycle, we first check if the product would have kicked out in Year 2. If so, it is recorded as a Year 2 early maturity and no further analysis is needed for that cycle. If not, we then check if it would have kicked out in Year 3 and so on through to Year 6.
We cannot select cycles starting later than 26th July 2007 as the six-year term would not have been completed. This is the case even when we know that on some of those occasions the product would have matured early. To include such cycles would be to “cherry pick” as onlythose that had matured early would be included; cycles where the product would not yet have matured and which could end in a capital loss would be left out and therefore would present a misleading result.
We can then analyse the outcomes and record how many of the 5971 cycles would have resulted in a Year 2 kick out, Year 3 kick out, Year 4 kick out, Year 5 kick out, Year 6 growth payment, a maturity payment equal to the capital invested or a capital loss at maturity.
The results can be presented as a percentage of the population of 5971 cycles. For example, there were 4710 Year 2 kick outs, which is 78.9% of the 5971 possible start dates; there were 56 occasions where a capital loss occurred – 0.9% of the possible start dates.
The same principles can be applied to other structures. For example, with the same basic product shape linked to the FTSE 100 and S&P 500 indices, we would start from the earliest common data date for both iShares ETF and the S&P 500 Index, which is 3rd January 1984. Although the time span is the same, there will be fewer observations, as the back-test can only use a start date when both indices are quoted. For example, a FTSE only product could start on 4th July but, as US markets are closed on 4th July, a FTSE/S&P product could not have done.
Our research is not restricted to fixed start dates however. It will facilitate a time period of the user’s choosing. For example, if an analyst were to assume that only the last 10 years of data were relevant, on our site they could set the start date to reflect only this period of data. The number of cycles would then be reduced to only the last 10 years.
Market Stress Test
Having only 1 sample of history constrains risk analysis: For example, some products would have had no, or a very small number of cycles that would have led to capital losses, despite the fact that the product structure has clear downside exposure and the fact that the investor is able to earn a significant additional potential return from assuming this risk. Placing such products in a low risk category may seem inappropriate.
Other products might have a structure that captures a current pricing opportunity to meet investor needs but which back-testing might show as having a poor historical record. Placing such products in a high risk category may seem equally inappropriate.
Put another way; past performance is not a reliable indicator of future performance and should not be used to assess future returns or risks. Nobel prize-winner Paul Samuelson (and many other leading economists) highlighted the fallacy in simply taking realised market returns (as the Historic Back-test does) as statistically significant proof of performance – statistically speaking, history is just one sample from a much larger distribution of possibilities.
Resampling the available data is a well-established way of representing the true underlying distribution of possibilities, enabling a statistically more correct assessment of risk. Resampling is a process that is simple, straightforward and robust.
As outlined in the Historic Back-test section above, there are 5971 dates on which a six year FTSE product might have started and reached the end of a six year term. This is just one dataset. Resampling allows for the creation of multiple datasets that explore more fully the range of possible movements, in this example by the FTSE 100 Index.
In order to determine the possible future movement in the underlying, we look at what the range of daily movements has been in the past for the index and use these to create new index series. On most days the Index does not move by large amounts – it may go up a little, go down a little or remain flat. On other days, the daily change might be more exaggerated and on comparatively fewer days there will be extreme movements.
The first step is to compile a new dataset from the information that we have. We can set the new intial ETF price base level at 100. A particular date is selected at random from all of the possible start dates in the data history that we are using, for example date 3214. The close of business level on date 3214 is recorded and compared to the level at close of business on day 3215. This gives us the movement for the Index on that day. Let’s say that it was 1% up.The Index level at the end of the first “day” of our new index is therefore 101 (i.e.100 increased by 1% or 100 x 1.01). This is repeated until we have a full dataset, the same size as the original dataset, so that we only ever create 5971 new cycles from the original 5971 cycles.
As the selection is random, any one-day return can be selected again, and so could be used more than once as part of the resampling process. Some daily returns may not get selected at all. We then carry out back-testing on the newly constructed dataset in exactly the same way as we did for the single historical back-test of the actual index performance.
We repeat this process multiple times until the results we get are stable. At this point we can be confident that the values that we are able to derive from the resampling process will represent the likely future range of returns of the underlying assets, that this range will remain constant, and so the results that we derive from the stress-test will not vary materially from one test to another. 750 repetitions of a 6,000 day price series is the point at which, our experience shows, the distribution of the return statistics stabilises i.e. the results after 750 repetitions is not materially altered from the results after 749 repetitions, further iterations would not bring about statistically meaningful changes in the results. This means the number of cycles tested rises to 4,500,000 (i.e. 6,000 * 750) and provides us with more data from which to calculate performance statistics of interest, for example ValueAtRisk. This much larger sample gives a more detailed view of the distribution of product outcomes and therefore enables more accurate estimates of performance statistics.
Market and Issuer Stress Test
We also calculate a set of risk and return numbers that reflect any additional risk that an investor is exposed to from possible issuer default.
The Credit Adjusted returns adjusts the risk and return numbers for the chance that the issuer defaults.
For each single cycle we assume that the final value is either
- The value that we have calculated using the re-sampled data
- 40% of the value calculated using the re-sampled data
The process is random, but the proportion of cycle results that are adjusted down will be based on the chance of default that we calculate. This process that we use implies that the issuer is as likely to default if the underlying assets have increased in value as when the underlying assets have fallen in value.
The chance of default is calculated using the issuer’s CDS levels, and an assumed 40% recovery rate. The 40% recovery rate assumes that in the event of issuer default investors will receive 40% of the value of the product when it matures. 40% is a market standard recovery rate that is typically used to calculate these figures; expected returns are not very sensitive to the choice of assumed recovery rate.
The credit adjustment will reduce the expected gains and reduce the chance of a gain since if there is a default, some of the instances where an investor would have made a profit will now be loss making. The credit adjustment will both increase the chance of a loss, and increase the expected loss. As a result the credit adjusted risk-return ratio and Win Lose ratios will both be lower.