This mail is the third in the series regarding the acceptance of RiskTrac VaR for UK Power. Please see the previous 2 days which will provide the history.

Essentially we have now reached the stage where we can explain the jump in VaR from moving from the VaR spread sheet to RiskTrac based largely on the good work that Fergus Trenholme - London Risk Management and Stig Faltinsen - London Research have done  (it has been a team effort as many others have been called in as appropriate and I would like to thank them too).

A full analysis is contained in the note from Fergus and Stig below but the key points to note are:

the changing of the input parameters partially explains the difference in the VAR number between the 2 systems but not completely (including the change in factor loadings)
the vast majority of the difference in the VAR numbers between the 2 systems is caused by the splitting of the volumetric position into the (primary and secondary) component curves in RiskTrac. In RiskTrac PPP is currently set as the primary curve even though 75% of the position is on the station gate curve where as in the spreadsheet the whole position is against the PPP curve

As a result of us now being able to isolate and explain the jump in VaR from moving to RiskTrac we will be reporting this number from now on. That said does the VaR number sound right ? The numer is theoretically but is the theory close to reality in the Uk Power market ?The RiskTrac calculated $25m VaR is a 1 day VaR and the  methodology would imply it should be broken 1 in 20 business days. Based on historical P&Ls this has not been the case and so it does not pass the "sense check". (note that the original $13m VaR was only broken once or twice in the last year).

There are many potential explanations why, but if we think about how the price and volatility curves move they typically only move at the short end (ie only the first 2 years) and the component VarR for this period is in the order of $8m. This could therefore imply that the volatilities we are using to calculate the VaR are too high. The book undoubtedly contains significant risk but due to market illiquidity more consideration should be given to "jumps" or "shocks".  The book does posses increased fat tail risk at this lower VaR level which is not  captured in the current VaR calculation.

Outstanding issues / work

Based on the second bullet point above consideration needs to be given to changing the primary and secondary curve settings as well as the correlations and factor loadings for and between all the curves. The impact of any such changes will need to be quantified. Fergus and Stig aim to work with Houston research and the business on this over the next few days / week(s) and will be providing a regular update to those on this mail list.

Additional work needs to done to assess whether the VaR  number  / methodology being calculated / employed by RiskTrac truly represents the expected volatility and performance of the portfolio and truly represents the 1 day earnings risk for 19 out of 20 days. It would also be appropriate, given the nature of the market and the term structure of the portfolio, to assess the fat tail risk being run.

I am away for the next 2 weeks so Fergus and Stig will be sending regular progress updates.

 
---------------------- Forwarded by James New/LON/ECT on 17/05/2001 19:10 ---------------------------


Stig Faltinsen@ENRON
17/05/2001 18:18
To:	James New/LON/ECT@ECT
cc:	Tani Nath/LON/ECT@ECT, Viacheslav Danilov/LON/ECT@ECT, Tanya Tamarchenko/Enron@EnronXGate, Kirstee Hewitt/LON/ECT@ECT, Fergus Trenholme/EU/Enron@Enron, Naveen Andrews/Enron@EnronXGate, Mike E Presley/ENRON@enronXgate 

Subject:	Analysis of variance between spreadsheet VaR and RisktRAC VaR for UK Power

Due to recent variances in UK Power VaR between the VaR spreadsheet (SS) model and RisktRAC (RT), despite small differences in Net Open Positions,  London Research and Risk Management have investigated these discrepancies with the results as below. 

COB 14 May 2001data was chosen as a test example 

			DPR			RT
NOP (TWh) 		43.460		42.255   
VaR	(USD MM)	14.118		24.749    

Summary of analysis: (for UK Power only)

Following checks of  parameter and position inputs to the London VaR SS, and a VaR recalc (with appropriate  inputs from RT), the resultant VaR change does not seem to explain the discrepancies between RT and the SS VaR number.
Consideration of the term structure of the Net Open Positions in both RT and SS showed only minor differences.
Adhoc VaR calculation using SS positions run through RT . The VaR obtained was comparable to the SS VaR. 
Adhoc VaR run through RT with curve positions correctly mapped to the relevant price curves, rather than aggregated to PPP curve.

Analysis of inputs to the London Spreadsheet (SS) VaR model

PLEASE NOTE THAT IN THE FIRST INSTANCE ALL CHANGES WERE PERFORMED ON A STAND-ALONE BASIS (i.e one change made and then returned to original state before changing another variable)

Correlations
Impact of RT correlations in SS : NEW VAR = USD 15.220 MM
(Energy Trend (ET) correlations mapped to PPPWD3, as RT does not contain ET correlation data. By keeping ET correlations the matrix became non positive definite, and hence unusable for VaR model). If the correlation matrix is larger (i.e. takes into account more markets) in RT as opposed to the SS this could also have a (marginal) effect (mainly because the Cholesky factorisation will be different). 


Factor loadings

FIVE VERSUS SEVEN FACTOR LOADINGS (SS uses 5 rather than 7 as per RT)

Houston Research (Jaesoo Lew) analysis has shown the effect of using five versus seven factor loadings. Considering one of the primary curves as an example  (PWD3). Five factor loadings describe 96.89 % of the movement of the curve whereas seven factors describe 98.53 %. Therefore the choice of whether to use 5 or 7 factors should not have a substantial impact on the VaR. 



Impact of RT factor loadings in SS : NEW VaR = USD 9.44 MM(ET (Energy Trends curve) with original factor loadings)
						   New VaR = USD  9.55 MM (ET factor loadings mapped to (PPP WD3))

Volatilities 
Impact of RT volatilities in SS : no change in VaR (vols are the same)

Prices 
Impact of RT prices in SS : no change in VaR (prices are the same with the impact of rounding errors providing very small immaterial differences)

Positions
Impact of RT positions in SS : no change in VaR

Forward forward volatilities 
Impact of utilising 'seasonality' flag (set to ON) in SS : NEW VaR = USD 14.650 MM
In RT these are assumed to be ON constantly, as are the fwd fwd volatilities (due to using implied rather than historic vol inputs)
Impact of utilising 'fwd fwd vol' flag (set to OFF) in SS : NEW VaR = USD 18.85 MM
The fwf fwd vol flag should be set to on in the SS model

Combining different parameters
The impact of RT correlations and RT factor loadings and several other combinations all gave changes to the VaR. However, the net result did not substantially differ from the results above (in blue).
 
Other issues:
Intramonth  positions -  not included in RT or  SS (for UKPower).  
Primary and secondary curve mappings. see below for Adhoc VaR results 
Currency conversion : Prices are in GBP. Does any term structure apply to GBP/USD FX (forward) curve used for transformation to USD VaR dollar Forward curve?

Therefore changing the input parameters partially explains the difference in the VaR between the two 'systems', but NOT completely. 

Checking differences in the term structure of positions

UK Power Portfolio is fed from three different sources; ET Elec positions are fed via a spreadsheet to the RT book "UKPowerswap1", The Eastern books are also fed via spreadsheet into RT books called "E1SB1, E1SB3, E2XX1,E2XX3". The UK Power book positions are uploaded directly from EnPower to the UK Power Portfolio in RT.
Differences:
ET ELEC : none
EnPower : total difference of (62.737) TWh (of which (63.700) is from June 01)
Eastern : total difference of (2,913,408) TWh (for the ' E1SB3' book only). Misfeed causing this error now corrected.

Ad-Hoc VaR Calculation Results

Run 1 :  Positions from the SS where ET Elec was mapped to WD3. VaR = USD 17.813 MM
Run 2 :  Positions from the SS where 1/12 of ET Elec positions were mapped to the 12 primary curves (PPP Wd and WE curves) 
VaR = USD 16.810 MM

The VaR results obtained from Runs 1 and 2 are comparable with the VaR number from the SS. The differences between these runs and the SS number can be attributed to parameter input variances (as described above)

Run 3 : Positions from SS split into component curves (i.e SG, N+P, NBP and PPP) rather than aggregated into PPP curves. ET Elec still split equally over the PPP positions. VaR USD = 25.432 MM.  This is almost exactly in line with the headline UKPower VaR calculated through RT.

UK Power exposure composition (TWh by curve price source)
PPP		(2.9)
NBP		9.5
N+P		(11.6)
SG		(32.6)
		_____
Total		(37.6)

Within the SS, all positions from above (i.e (37.6) TWh) were 'mapped' to the PPP curves, whereas in RT all curves were mapped individually as secondary curves (where PPP acts as primary curve) - even though PPP curve has smallest amount of price risk attached to it through having the smallest Net Open Position. The adhoc results support the assumption that  most of the RT VaR within the PPP_NEW book is due to SG and N+B rather than PPP.

The business may now want to consider the primary / secondary relationships used within RT for UKPower based upon the net exposures against each curve. Research suggests that SG should now become a primary curve based upon these results if price data is available for factor loading and correlation data analysis.. Further analysis on the impact of this change would be needed to quantify the effect of this.

As VaR (at the 95% confidence interval) should typically represent expected losses over 1 day for 1 day in a 20 day period, backtesting should demonstrate this by illustrating a realised PnL loss greater than the VaR number. However, for UK Power, the realised PnL losses have only exceeded the VaR approximately once over the last 300 days. By increasing the reported VaR number, this may not seem appropriate given the underlying reasoning behind VaR and empirical evidence from backtesting. It should be borne in mind that the current structure of the UK Power markets, since the introduction of NETA, has only been in operation for approximately 45 days. For this reason, all data considered for either correlation or factor loading analysis should be considered with care. 

The introduction of NETA into the UK Power market has necesserily created more risk to Enron, especially until the market settles down and gains maturity and stability. Therefore, the revised UK Power VaR of USD 25 MM is not entirely inappropriate, although still appears a high number in absolute terms. Research and Risk Management will investigate the underlying parameter levels during the next few days now that the core assumptions have been verified.