Related to the uncertainty of a banks earnings on its trading portfolio caused by changes in market conditions such as:

The price of an asset Interest rates Market volatility Market liquidity

Debate: is spread risk a part of market risk or of credit risk? JP Morgan classifies spread risk as credit risk Market risk used by bank managers and regulators is not synonymous with systematic market risk analyzed by investors in securities markets, the latter reflecting the comovement of a security with the market portfolio

 It emphasizes the risks to banks that actively trade assets and liabilities, including derivatives, rather than hold them for LT investment, funding or hedging purposes. Income from trading activities is increasingly replacing income from the traditional banking activities of deposit taking and lending The resulting earnings uncertainty can be measured over periods as short as a day or as long as a year. In absolute terms, market risk can be defined as a dollar exposure amount or as a relative amount against some benchmark

 It is about losses in on and off-BS positions arising from movements in market prices; risks subject to this requirement are:
Risks pertaining to interest rate related instruments and equities in the trading book Foreign exchange risk and commodities risk throughout the bank

1995 proposal: application of capital charges to the current market value of open positions (including derivatives positions in interest rate related instruments and equities in banks trading books, and to the banks total currency and commodities positions

Assets

Liabilities

Loans
Banking book

Capital

Other illiquid assets Deposits Bonds (long) Bonds (short)

Commodities (long) Commodities (short) Trading book FX (long) Equities (long) Derivatives (long) FX (short) Equities (short) Derivatives (short)

 The assessment of capital charges is achieved through the use of two alternative techniques: The standardized approach allows measurement of the four risks:
Interest rate Equity position Foreign exchange Commodity risks

by using sets of forfeits, for defining to which base they apply, allowing some offsetting effects within portfolio of traded instruments; these effects reduce the base for calculating the capital charge by using a net exposure rather gross exposures

 The proprietary model approach allows banks to use risk measures derived from their own internal risk management models, subject to a number of conditions, related to qualitative standards of models and processes 1995 proposal allows banks to use Tier 3 capital, essentially made up of ST subordinated debt to meet their market risks, which is subject to a number of conditions, such as being limited to market risk capital and being subject to a lock-in clause, stipulating that no such capital can be repaid if that payment results in a banks overall capital being lower than a min capital requirement

 The standalone market risk of an instrument is a VaR (or max potential loss) resulting from P/L distribution derived from the underlying market parameter variations Thing become more complicated for portfolios because risks offset to some extent; the portfolio effect reduces portfolio risk, making it lower than the sum of all the individual standalone risks, eg LT & ST exposures to the same instrument are exactly offset; also LT & ST exposures to similar instruments are partially offset (the price movements of a 4 and a 5 year bond correlate because the 4 and 5 year interest rates do). Still, this correlation is not perfect since a 1% change in the former does not mechanically trigger a 1% change in the latter

 With different instruments, eg equities and bonds, a conservative view would consider that all prices move adversely simultaneously; this makes no sense because the market parameters (interest rates, foreign exchange rates and equity indexes) are interdependent, some tending to move in the same direction, and others in opposite directions. The strength of such associations varies across pairs of market parameters, and the statistical measure for such associations is the correlation: market parameters comply with a correlation structure, so that the assumption of simultaneous adverse co-movements is irrelevant

 This leads to the basic concepts of general vs specific risk. General risk: dependent on market parameters driving all prices, this dependence generating price co-movements, or price correlations, because all prices depend, to some extent, on a common set of market parameters Specific risk: it is the price volatility unrelated to market parameters. It designates price variations unrelated to market parameters variations. Statistically, it is easy to isolate specific risk over a given period. It is sufficient to relate prices to market parameters statistically to get a direct measure of general risk; the remaining fraction of the price volatility is the specific risk

 Basel Committee: 2 approaches, including the following critical inputs:

The current valuation of exposures Their sensitivities (variations of values for a unit change in market parameter) to the underlying market parameters Rules governing offsetting effects between opposing exposures in similar instruments and, eventually, across the entire range of instruments

Full-blown models should be able to capture all valuation, sensitivity and correlation effects

 Standardized approach: reliance on forfeits for assessing capital charges as a % of current exposures, on grids for capturing the differences in sensitivities of various market instruments and on offsetting rules allowing netting of the risks within a portfolio whenever there is no residual basis risk. Forfeits capture conservatively the potential adverse deviations of instruments. They vary across market compartments and products, such as bonds of different maturities, because their sensitivity varies. Stocks and bonds carry forfeits of 8% of net balances

 The addition of individual risks, without offsetting exposures, overestimates the portfolio risk because the underlying assumption is that all adverse deviations occur simultaneously; with a given class of instruments, such as bonds, equity or foreign exchange, regulators allow offsetting risks to remain to a certain extent, eg being long and short on the same stock results in zero risk, because the gain in the long leg offsets the loss in the short leg when the stock goes up, and vice versa. Offsetting is limited to exact matches of instrument characteristics In other cases, regulators rely on the specific vs general risk distinction, following the principle of adding specific risk forfeits while allowing limiting offsetting effects for general risk, the rationale being that general risk refers to co-movement of prices, while specific risk is unrelated to underlying market parameters

 For interest rate instruments, the capital charge adds individual transaction forfeits, varying from 0% (government), to 8% for maturities over 24 months Offsetting specific risks is not feasible except for identical debt issues The capital requirement for general risk captures the risk of loss arising from charges in market interest rates There are 2 methods for measuring general risk:
A maturity method A duration method

 Maturity method: it uses 13 time bands for assigning instruments, aiming at capturing the varying sensitivities across time bands of instruments. Offsetting long and short positions within time bands is possible, but there are additional residual capital charges applying because of intra-band maturity mismatches; since interest rates across time bands correlate as well, partial offsets across time bands are possible. This procedure uses zones of maturity buckets grouping the narrower time bands. Offsets decrease with the distance between the time bands. The accord sets %s of exposures significantly <1, to cap the exposures allowed to offset (100% implying a total offset of the entire exposure)

 Duration method: it allows direct measurement of the sensitivities, skipping the time band complexity, by using the continuous spectrum of durations. It is necessary to assign sensitivities to a duration-based ladder, and within each slot, the capital charge is a forfeit. Sensitivities in values should refer to preset charges of interest rates, whose values lie between the 0.6% to the 1.00% range. Offsetting is subject to a floor for residual risk (basis risk, since there are duration mismatches within bands)

 Derivatives: combinations of underlying exposures, and each component is subject to a floor for residual risk, eg a swap is a combination of 2 exposures, the receiving leg and the paying leg; thus, forfeit values add unless the absence of any residual risk allows full offsets There is no specific risk for derivatives. For options, there are special provisions based on sensitivities to the various factors that influence their prices, and their charges with these factors. The Accord proposes the scenario matrix analysis, where each cell of the matrix shows the portfolio value given a combination of scenarios of underlying asset price and volatility. The highest loss value in the matrix provides the capital charge for the entire portfolio of options

 Equities: application of the basic distinction between specific vs general risk: the capital charge for specific risk is 8%, unless the portfolio is both liquid and well diversified, in which case the charge is 4% Offsetting long and short exposures is feasible for general risk, but not for specific risk The general market risk capital charge is 8% Equity derivatives should follow the same decomposition rule as underlying exposures

 Foreign exchange: 2 processes serve for calculating the capital requirement for foreign exchange risk
The 1st is to measure the exposure to a single currency position The 2nd is to measure the risks inherent in a banks mix of long and short positions in different currencies

Structural positions are exempt of capital charges, referring to exposures protecting the bank from movements of exchange rates, such as assets and liabilities in foreign currencies remaining matched The shorthand method treat all currencies alike, and applies the 8% capital charge over the greatest netted exposure, either long or short

 Commodity risk is more complex than market instrument risk because it combines a pure commodity price risk with other risks, eg basis risk (mismatch of prices of similar commodities), interest rate risk (for carrying cost of exposures) and forward price risk Additionally, there is directional risk in commodities prices The principle is to:
Assign transactions to maturity buckets, Allow offsetting of matched exposures, and Assign a higher capital charge for risk

Principles of extending to proprietary market risk models: Market risk is the risk of loss during the min period required to liquidate transactions in the market The potential loss is the 99th loss percentile (one-tailed) for market risk models This min period depends on the type of products, and on their sensitivities to a given variation of their underlying market parameters. In general, a 10 day period for liquidating position is the normal reference for measuring downside risk Potential losses depend on market movements during this period, and the sensitivities of different assets

 Models should incorporate historical observation over at least 1 year. Additionally, the capital charge is the higher of the previous days VaR, or 3 times the average daily VaR of the preceding 60 days A multiplication factor applies to this modeled VaR, accounting for potential weaknesses in the modeling process or exceptional circumstances. Additionally, regulators emphasize:
Stress-testing, to see what happens under exceptional conditions, using extreme scenarios maximizing losses to find out how large they can be Back-testing of models to ensure that models capture the actual deviations of the portfolios, using the model with past data to check whether the modeled deviations of values are in line or not with the historical deviations of values

BASEL II min standard for the use of VaR to calculate market risk for the assignment of regulatory capital adequacy VaR must be computed on a daily basis, using a 99th percentile, one-tailed confidence interval A min holding period of 10 trading days must be used to simulate liquidity issues that last for longer than the 1-day VaR holding period (the square root of time may be applied to the 1-day VaR estimate, however, to simplify the calculation of this VaR measure) A min of a 1-business year observation period (250 days) must be used, with updates of data sets taking place every day, and reassessments of weights and other market data should take place no less than once every 3 months

 Banks are allowed discretion in recognizing empirical correlations within broad risk categories, eg interest rates, exchange rates, equity prices and commodity prices Banks should hold capital equivalent to the higher of either the last days VaR measure or an average of the last 60 days (applying a multiplication factor of at least 3) The banks VaR measure should meet a certain level of accuracy upon back-testing, or else a penal rate will be applied to its charges, eg a plus factor. If the model fails 3 consecutive times, the banks trading license may be revoked

THE RISKMETRICS MODEL The bank is concerned with how much it can potentially lose if market conditions move adversely tomorrow, thus:
Market risk = Estimated potential loss under adverse circumstances

Market risk is measured in terms of the banks daily earnings at risk (DEAR, equal to market risk exposure over the next 24 hours), consisting of 3 components
DEAR = ($ market value of the position) X (price sensitivity of the position) X (potential adverse move in yield), or DEAR = ($ market value of the position) X price volatility)

The Riskmetrics model calculates DEAR in 3 trading areas, fixed income, foreign exchange (FX) and equities, and then it estimates the aggregate risk of the entire trading portfolio

Fixed income securities A bank has a $1 million market value position in zerocoupon bonds of 7 years to maturity with a face value of $1,631,483. Todays yield on these bonds is 7.243% p.a. These bonds are held as part of the trading portfolio, thus:
$ market value of the position = $1 million

The potential loss depends on the bonds price volatility, where

Daily price volatility = (price sensitivity to a small change in yield) X (adverse daily yield move) = (MD) X (adverse daily yield move)

The modified duration (MD) of this bond is:

MD = D / 1 + R = 7 / (1.07243) = 6.527, given that the yield of the bond is R= 7.243%

To estimate price volatility, multiply the bonds MD X the expected adverse daily yield move

 This analysis can be extended to calculate potential losses over 2, 3, N days. Assuming that yield shocks are independent and daily volatility is approximately constant, and the bank is locked in to holding this asset for N number of days, then the N-day market value at risk (VAR) is related to DEAR by: VAR = DEAR X N That is, the earnings a bank has at risk, should interest rate yields move against the bank, are a function of DEAR for 1 day and the square root of the number of days that the bank is forced to hold the securities because of an illiquid market. In specific, DEAR assumes that the bank can sell all the bonds tomorrow, even at the new lower price; in real terms, it may take many days for the bank to unload its position

FX A bank had a $1.6 million trading position in spot euros at the close of a business day. The bank wants to calculate the DEAR from this position (the risk exposure on this position should the next day be a bad one in the FX markets with respect to the value of the against the $ First we calculate the $ value of the position:
$ equivalent value of the position = (FX position) X (/$ spot exchange rate) = ($1.6 million) X ($ p.u. of foreign currency)

If the exchange rate is 1.60/$1 at the daily close, then:

$ value of the position = ($1.6 million) X (0.625/) = $1 million

 If we look back at the daily changes in the /$ exchange rate over the past year, we find that the volatility, or standard deviation () of daily changes in the spot exchange rate was 56.5 bp. If the bank is interested in adverse moves (bad moves that will not occur more than 5% of the time, or 1 day in every 20). Statistically, if changes in exchange rates are historically normally distributed, the exchange rate must change in the adverse direction by 1.65 (1.65 X 56.5 bp), for this change to be viewed as likely to occur only 1 day in every 20 days, thus:
FX volatility = 1.65 X 56.5 bp = 93.2 bp or 0.932%

In other word, during the last year, the declined in value against the $ by 93.2 bp 5% of the time. As a result:
DEAR = ($ value of the position) (FX volatility) = ($1 million) X (.00932) = $9,320

Equities From the Capital Asset Pricing Model (CAPM) we know that there are 2 types of risk to an equity position in an individual stock i (assuming that both of them are independent of each other):
Total risk = Systematic risk + Unsystematic risk, or (2it) = (2i 2mt) + (2eit)

Systematic risk reflects the comovement of that stock with the market portfolio reflected by the stocks beta (i, systematic risk reflecting the comovement of the returns on a specific stock with returns on the market portfolio), and the volatility of the market portfolio (mt), while unsystematic risk is specific to the firm itself (eit)

 In a very well diversified portfolio, unsystematic risk can be largely diversified, that is it = 0, leaving behind systematic (undiversifiable) market risk. If the banks trading portfolio follows (replicates) the returns on the stock market index, the of the portfolio will be 1, since the movement of returns on the banks portfolio will be 1 to 1 with the market, and the standard deviation of the portfolio (it), will be equal to the standard deviation of the stock market index (mt) If 1, as in the case of most individual stocks, then:
DEAR = $ value of the position i 1.65, where i is the systematic risk on the ith stock

 In less well diversified portfolios of individual stocks, the effect of unsystematic risk (eit) on the value of the trading position would need to be added. Furthermore, if the CAPM does not offer a good explanation of assets pricing compared with, eg multi-index arbitrage pricing theory (APT), a degree of error will be built into the calculation of DEAR Derivatives are also used for trading purposes. In calculating a derivatives DEAR, the derivative has to be converted into a position in the underlying asset, eg bond, FX or equity)

Criticisms of Riskmetrics The need to assume a symmetric (normal) distribution for all asset returns: for some assets, such as options and ST securities, this is highly questionable, eg the most an investor can lose if he/she buys a call option on an equity is the call premium; however, the investor's potential upside returns are unlimited. In a statistical sense, the returns on call options are nonnormal since they exhibit a positive skew (for a normal distribution, its skew, being the 3rd moment of the distribution) = 0)

 It ignores the (risk in the) payments of accrued interest on a banks debt securities, thus VAR models will underestimate the true probability of default and the appropriate level of capital to be held against this risk Because of the distributional assumptions, while Riskmetrics produces reasonable estimates of downside risk for banks with highly diversified portfolios, banks with small, undiversified portfolios will significantly underestimate their true risk exposure using Riskmetrics Many asset distributions have fat tails, thus Riskmetrics, by assuming the normal distribution, underestimates the risk of extreme losses

 VAR models by definition concern themselves with risk rather than return, however, minimizing risk may be highly costly in terms of the return that the bank give up. Indeed, there may be many more return-risk combinations preferable achieved at the min risk point in the trading portfolio. Recent upgrades to Riskmetrics allow management to incorporate a return dimension to VAR analysis so that management can now evaluate how trading portfolio returns differ as VAR changes

Historical (back simulation) approach. Advantages:

Simplicity Does not require that asset returns be normally distributed Does not require that the correlations or standard deviations of asset returns be calculated

Essential idea: take the current market portfolio of assets (FX, bond, equities etc) and revalue them on the basis of the actual prices (returns) that existed on those assets yesterday, the day before that and so on. A bank will calculate the market or value risk of its current portfolio on the basis of prices (returns) that existed for those assets on each of the last 500 days; then it will calculate the 5% worst case (the portfolio value that has the 25th lowest value out of 500, or, in other words only 25 days out of 500, 5% of the time, would the value of the portfolio fall below this number based on recent historic experience of FX rates, equity price changes, interest rate changes and so on

 Step 1: measure exposures, by converting todays foreign currency positions into $ equivalents using todays exchange rates Step 2: measure sensitivity of each FX position by calculating its delta, where delta measures the change in the $ value of each FX position if the or the SF depreciates by 1% against the $ (in the case of FX delta measures the $ change in FX holdings for a 1% change in the FX rate; in the case of equities, it measures the change in the value of those securities for a 1% change in price, while for bonds it measures the change in the value for a 1% change in the price of the bond; note that delta measures sensitivity of a bonds value to a change in yield, not price)

 Step 3: measure risk, by looking at the actual % changes in exchange rates, /$ and SF/$, on each of the past 500 days Step 4: repeat Step 3, calculating FX losses and/or gains on each of the past 500 trading days, excluding weekends and holidays, when the FX market is closed. This amounts going back in time over to 2 years. For each of these days we calculate the actual change in exchange rates and multiplied by the deltas pf each position; then, these two numbers are summed to attain total risk measures for each of the past 500 days

 Step 5: rank days by risk from worst to best; these risk measures can then be ranked from worst to best. We are interested in the 5% worst case, that is, a loss that does not occur more than 25 days out of 500 Step 6: VAR. if it is assumed that the recent past distribution of exchange rates is an accurate reflection of the likely distribution of FX rate changes in the future (that exchange rate changes have a stationary distribution), then the worst-case loss can be viewed as the FX VAR exposure on the bank, with a 5% probability of this case emerging tomorrow

The main disadvantage of the historic (back simulation) approach, is the degree of confidence we have in the 5% VAR number based on 500 observations. Statistically speaking, 500 observations are not very many, so there will be a very wide confidence band (or standard error) around the estimated number. If we attempt to solve this problem by go back in time more than 500 days and estimate the 5% VAR based on 1,000 past daily observations (the 50th worst case) or even 10,000 past daily observations (the 500th worst case), past observations may become decreasingly relevant in predicting VAR in the future

Monte Carlo simulation approach Returns or rates generated reflect the probability with which they have occurred in recent historic time periods Step 1: calculate the historic variance covariance matrix () of FX changes Step 2: decompose this matrix into two symmetric matrices, A and A (the only difference between them being that the number in the rows of A becomes the number in the columns of A, thus = AA), allowing the bank to generate scenarios for the FX position by multiplying the A matrix, which reflects the historic volatilities and correlations among FX rates, by a random number vector z (assumed to be normally distributed with a mean of 0 and a standard deviation of 1 or z ~ N (0, 1)

 Step 2 (continued): thus, 10,000 random values of z are drawn for each FX rate, eg:
If y is an FX scenario, then y = Az. For each FX rate, 10,000 values of z are randomly generated to produce 10,000 values of y; these y values are then used to revalue the FX position and calculate losses and/or gains

Step 3: calculate VAR similarly to the historic approach, only in the Monte Carlo approach the VAR is the 500th worst simulated loss out of 10,000