RiskAPI has historically provided multiple methods for accessing continuous, back-adjusted futures contract data. These include multiple notations for creating complete time series composed of mathematically combined, discrete contract data sets. Users have had the option to back-adjust this data using arithmetic or proportional roll-date price differences.

Continuous futures contracts combine all intra-year individual contract time series data into a single, uninterrupted set of data, allowing users to perform analysis on long term histories that do not exist due to a lack of trading data for individual monthly contracts. Data from each individual contract is mathematically combined, with each series being combined at a "roll-date" preceding the expiration date of the contract.

For most speculators, these notations work well since they rarely hold a physical commodity contract to expiration. Doing so would result in either a) accepting delivery of the underlying physical commodity (long position) or b) facilitating the delivery of the underlying physical commodity (short position). However, for physical traders and real hedgers, holding a contract to expiration is desirable and is done explicitly to take advantage of the delivery features of a physical commodity contract. Typically, periods just prior to and up-to physical commodity futures expiration contain higher volatility due to the price and liquidity dynamics involved in the expiration/delivery process. Risk managers of physical commodity trades therefore wish to capture this additional volatility in their analysis.

RiskAPI now includes two additional futures contract symbol notations allowing users to employ continuous contract data with discrete contracts joined (or "rolled") at each contract expiration date. The resulting continuous data set, therefore, also incorporates the unique price dynamics inherent to delivery and expiration.

Continuous contracts rolled at expiration can be accessed by a simple combination of an "E1-E20" suffix or "P1-P20" suffix. For example, the symbol "CLE3" represents the 3rd continuous CME WTI Crude Oil contract, with individual monthly contract data joined and rolled at each respective contract expiration date. To avoid introducing artificial volatility due to price differences at the roll dates, data is back adjusted using arithmetic differences between each contract at the roll dates. Using the format "CLP3" instead will produce an analogous data set, with the back-adjustment using proportional (i.e. percent) differences between the contracts at each roll date.

With winter in the northern hemisphere approaching, temperatures are cooling and those looking to hedge exposure to the energy markets keenly observe the behavior of both the Henry Hub Natural Gas and WTI Crude Oil curves

Using the RiskAPI Add-In, a one-year, daily correlation matrix is generated on the NG curve, starting at the current front month and going out a full calendar year:

 

 

Similarly, a daily correlation matrix for the WTI curve is also generated:

 

 

Note that values shaded in green show larger correlation and values shaded in red show smaller correlation.

As winter approaches in the northern hemisphere, the potential for outsized moves in global natural gas markets increases. These can take place due to either unexpected cooling or warming periods which may significantly affect consumer demand and produce large changes in natural gas prices.

In general, price dynamics in natural gas markets are uniquely challenging for portfolio and risk managers. This is partly because of the difficulty in forecasting weather but also since natural gas volatility tends to increase quite dramatically as actively trading futures contracts become "prompt" (i.e., as they approach their delivery dates). The historical returns for a given contract may be well-behaved and adhere to expected volatility estimates throughout the contract's trading life. However, as the delivery date of the contract approaches and it becomes prompt, volatility tends to explode higher, rendering any historical-data driven analysis quite unpredictive and falling well short of correctly estimating the magnitude of future returns.

This effect on volatility is demonstrated quite well by the behavior of the most recent front contract for the NYMEX Henry Hub Natural Gas future. Henry Hub Natural Gas is a monthly futures contract that expires in the final trading days prior to each delivery month. There are 12 contracts that trade for each calendar year, each with its own delivery date at or near contract expiry. This highly liquid contract trades several years out into the future, providing a very efficient price discovery, speculation, and hedging mechanism for the US natural gas markets. The December 2020 future, for example, was the prompt contract during the entire month of November 2020:

 

 

As can be seen above, the realized volatility of this contract remains fairly range-bound for most of the calendar year 2020. However, as the contract approaches the prompt month (expiration at the end of November), we see an effective doubling of volatility, from 21% in September to almost 40% by contract expiration on November 25th, 2020. Applying a statistical estimate such as Value at Risk (VaR) to this contract's data would vastly underestimate exposure prior to the prompt period. The problem mainly arises from the desire to include more data in a given statistical analysis, yet when included, this data has the effect of underestimating future exposure since it mostly contains returns observations from a vastly lower risk regime (i.e. all the historical returns preceding the prompt period).

The effect is greatly pronounced when looking at VaR during the prompt month itself, specifically during two dates: November 3rd and November 16th. Two commonly used VaR models, Parametric (also known as delta-normal) and Historical Simulation, both vastly underestimate actual return magnitudes on both of these days:

 

 

Here we see traditional approaches to VaR performing quite badly, with 99% confidence, 1-day VaR estimated using 1 year of data up to and before each date (t-1), falling well short of the actual P&L for a single long contract of NGZ20. 99% Confidence Parametric VaR as of 11/2/2020, for example, estimates a maximum loss of $1,1920, (a 3.18% return), while the actual loss on 11/3 was $1,850 (or -5.70%). The outsized loss on 11/16 far exceeds both methods of risk estimation, with Historical Simulation VaR, even as the better of the two methods, still coming short by more than a factor of 2.

As a remedy to this problem, RiskAPI includes Volatility-Adjusted Futures (VAF's), a class of mathematically constructed futures contract data that utilizes historical returns combined with measurements of recent volatility to normalize past returns to current volatility conditions. The idea behind this construct is to preserve the statistical significance of historical returns, while making them relevant to present day market conditions, specifically with regards to current volatility. In short, the construct allows users to take advantage of large amounts of historical return observations and to simulate these returns behaving as if they took place under similar, present day volatility conditions.

The RiskAPI system provides a simple format for converting any futures contract to use this volatility sampling and normalization process in order to seamlessly adjust historical returns. For the December 2020 Henry Hub Natural Gas contract, symbol: NGZ20, a symbol modification is applied by adding a hash sign to this code: NGZ20#. This prompts the system to modify the historical returns of this contract by normalizing historical return observations to present day volatility conditions using past and present volatility measurements.

Using the RiskAPI Add-In, which provides a simple keyword-based mechanism to quickly generate VaR in Excel via the RiskAPI cloud service, we reproduce the same calculations, this time on the VAF version of the Henry Hub contract, NGZ20#:

 

 

The result is a much better-performing estimate of losses, using only data preceding each outsized return event:

 

 

Additional manipulations of returns are available, allowing users to take advantage of filtered, seasonal samples, i.e., constructing a set of normalized returns sampled from specific months out of each year, going back several years (a full whitepaper of this methodology exists for registered and trial users of the system).