FRM P1 Edgy Rhinx 27-09-2021 1 Enron Disasters corp. governance, Arthur Andersen lost license to SEC, Northern Rock repo financing, funding JPM and City were liquidity SG counterparties, Agency risk, Jerome Kerviel, fake offsetting SOX 2002 transactions UBS incorrect modelling of long-dated options Drysdale collateral Kidder Jett, artificial, sell forward Barings 1995 Leeson, long-long Nikkey Allied Irish Rusnak, FX, fictioneous trades MGMR 1993 stack and roll oil, f. World.com corp. governance Global Crossing corp. governance SWIFT 2016 Bangladesh bank , 81m Volkswagen 2015 emission scandal Nabisco lev. buyout, spread 100 to 350bp MF Global CEO on the Board invested in EU during Greek crisis liquidity, backward. → contango LTCM stressed correlations, f. liquidity, shorted govt. US and 2 Germany vs flight to quality Bankers Trust P&G, Gibson, complex Orange County 1994 R. Citron, complex CML: E(Rp ) = r + Cont. Illinois MBS and real estate f. by 1d Rm −r σm i σp T P Im = Rm − r E(RI ) repo βi = b = rapid growth during 1970 Jensen α = E(Rp ) − [r + βp (Rm − r)] p E(Rp )−E(Rb ) IR = ε = V ar(Rp − Rb ) ε q P n ε = n1 t=1 (Rp,t − Rb,t )2 = p E[(Rp − Rb )2 ] − (E[Rp − Rb ])2 boom, in 1984 Penn Square failed → loss of confidence Niederhoffer 1997 lost on naked puts JPM 2012 Bruno Iksil, London Cov(i,m) 2 σm E(R )−R p min Sortino = q P σ1/2 n σ1/2 = n1 t=1 [(Rp,t − Rmin )− ]2 whale, corp. governance, Sachsen h E(Rp )−r Sharp = σp E(Rp )−r T reynor = βp E(Rp )−r βp = Rm −r , Rm ≈ inverse swaps Lehman CAPM complex cred. derivatives Fama and French 2007 Landesbank, leveraged E[Rp ] = α + β1 (Rm − r) + β2 SM B + β3 HM L + AAA, CEO Ken Law also β4 RM W + β5 CM A, chaired 1 size small – big exhaustive capitalization high – low P (A|C) × P (B|C) = P (A ∩ B|C) – conditionally operational PL robust – weak independent, taller students do better given age 8 investments conservative – agressive TPI > Market > Sharp: positive α, lacks 40% of (blue and red) are convertibles gives no diversification. Conditional E[x] is E[x] weighted by conditional additional info about blue and red events. probabilities. 3 Funds Open-end funds – not on exchange, but sold at 5 NAV, transactions at 16:00, no stop-orders σp2 = w12 σ12 + w22 σ22 + 2w1 w2 p12 σ1 σ2 (unknown price). Covs (X, Y ) = βs2X = Closed-end – on exchange, can be shorted, below Covariance n n−1 Cov(X, Y ) = RsX sY n Covs (X, Y ) = n−1 [E(X, Y ) − E(X)E(Y )] = hP i (XY ) n − xy n−1 n NAV. ETF – on exchange, shorted, at NAV, disclosed Spearman’s rank correlation twice a day. Kendal’s τ = Undesirable Late trading, after 16:00 Market timing, stale stocks nc −nd n(n−1)/2 = nc −nd nc +nd +nt nc concordant Xi > Xj → Yi > Yj nd discordant Xi > Xj → Yi < Yj nt ties Xi = Xj → Yi = Yj Pn Sum of mean deviations is 0: i=1 (xi − µ̂) = 0 Pn 0] = 12 (nσ 2 ) = Cov V ar(µ̂) = n12 [ i=1 V ar(Xi ) + n Front running Direct brokerage Defences σ2 n Hurdle rate High watermark clause Minimum Variance Portfolio Clawback clause 2 σ12 = w2 σ12 + (1 − w)2 σ22 + 2w(1 − w)σ12 , Backfill bias – hedgefund reports results when w= σ22 −σ12 σ12 −2σ12 +σ22 6 Hypothesis profitable. Incentive Fees H0 : x = x0 , x ≤ x0 , x ≥ x0 √ √ [µ − tα σ/ n, µ + tα σ/ n] 2 + 20% Prior Post 0.8(R − 0.02), R > 0.02 P (T ype I, rejected true H0 ) = α = P (Crit. value) 0.8R − 0.02(1 + R), R > 0 P (T ype II, f ailure to reject f alse H0 ) = β With probabilities Significance level = α (5%) Overall = RP (R) − LP (L) 95% = 1 − α, degree of confidence To fund = 0.02 + 0.2P (R)(R − 0.02) Power of test = 1 − β To investor = 0.8P (R)(R − 0.02) + P (L)(−L − 0.02) 4 Equality of two means (vs Normal) T = Probability µz σ √z n = µx −µy r x−µ √ , s/ n 2 +σ 2 −2Cov(x,y) σx y n P (A ∪ B) = P (A) + P (B) − P (A ∩ B) tstat = P (A ∩ B) = P (A|B)P (B) −−−→ P (A)P (B) (size < 30 and Normal) P (B|U ) = P (U |B)P (B) P (U ) = p-value = P (T eststat ) indp. P (U ∩B) P (U ) c −→ iid µ −µy r x 2 σx nx σ2 + nyy when (V unknown and size ≥ 30) or zstat , when (V known and Normal) or (V unknown c P (U ) = P (U |A)P (A) + P (U |A )P (A ) = and n is large) P (U ∩ A) + P (U ∩ Ac ) – mutually exclusive and χ2 = 2 (x−1)s2 σ02 F = s2big 2 ssmall Survivorship N (−1) 0.159 Simultaneity N (−1.28) 0.1 Omitted variable N (−1.645) 0.05 Attenuation or fading, measured with ε, N (−1.96) 0.025 N (−2.33) 0.01 N (−2.58) √ εµ = σ/ n leads to underestimation of regression coefficients Omitted variable: βˆ1 = β1 + β2 ρ12 σ1 σ2 Cook’s Distance > 1 → outlier Pn (Yˆj )2 Dj = i=1 ksi 2 > 1, 0.005 √ εσ = σ/ 2n f k – n of variables s2f – squared residuals of the full model 7 β= Regression Cov(X,Y ) V ar(X) Estimator properties α = Y − βX 2 R2 ∈ (−∞, 1] = rXY = rY2 Ŷ for multiple P ESS = (Ŷ − Y )2 P 2 T SS = (Y − Y ) = + P RSS = (Y − Ŷ )2 1= ESS+RSS T SS RSS T SS 2 Adjusted R = 1 − = 1 − R2 RSS T SS × Unbiased expected val. = parameter, E(µ̂) − µ = 0 Efficient best possible, least V ar(ε) estimation Consistent sample size increases, bias goes to zero Linear Transformation Properties n−1 n−k−1 Y = a + bX Dtstat = β−0 ε ESS/k Fall = RSS/(n−k−1) , n − k − 1 degrees of freedom (RSS −RSS )/q Fpartial = RSSfp/(n−kff−1) , q – number of removed E(Y ) = a + bE(X) 2 V (Y ) = b2 σX Cov[a + bX1 , c + dX2 ] = bd Cov[X1 , X2 ] variables (restrictions) p SEE = RSS/(n − k − 1) SY = sign(b)SX KY = KX Ft > Fcrit full model contributes meaningfully High F → superior fit → rejects H0 : βi = 0 F rejects multicollinearity, P (T ype 2 error) ↑ T does not ≈ 0 No individual variables effects, but common source 8 Cov. stationary if the first 2 moments {µ, Cov(Xt , Xt−1 )} are stationary explains. F does not Forecasting, Stationarity PACF – partial autocorr. function of Yt and Yt−h Yule-Walker equations (YW) estimate autocorr. variable has small effect T rejects Gauss-Markov theorem: OLS is BLUE, ε ∼ N (iid), function yt = Φ|t| y0 , no outliers, E(ε) = 0 → relax normality assumption √ 2 σ x = σy = 1 → b = R σ xy β = σ2 σ 2 σxy x y σxy σx2 α = Y − βX Violations Heteroskedasticity σε2 , OLS → W LS Multicollinearity Corr(β1 , β2 ) > 0.9, V IF > 10 t = 0, 1, 2, .. 0.651 , 0.652 , .. y1 = ϕ1 y0 + ϕ2 y−1 ρ1 = y1 /y0 = ϕ1 1−ϕ2 E(Yt ) = d + ϕYt−1 + ϵt , P stationary |ϕ| < 1, absolute values of slope coefficients. long-run µ = d 1−ϕ , d – intercept Serial correlation of ε 1 Variance inflation V IF = 1−R 2 , omit the variable V (Yt ) = to cure multicollinearity MA is covar. stationary Biases Function cuts off at order of the process i 3 σε2 1−ϕ2 ACF → it’s MA process Unit root PACF → it’s AR (seasonality) Yt = Yt−1 + ε both no good fit Box-Pierce and Ljung–Box tests Ph Qbp = n i=1 ρ2i , h – number of lags, ρ – autocorrelation Ph Ph 2 Qlb = n i=1 n+2 i=1 n−i ρi = n(n + 2) Qc ∼ χ21−α,h , Yt = Y0 + ε1 + ε2 + .. + εt (1 − L)(1 − 0.65L)Yt = εt 9 ρ2i n−i Sequences MA(1) Yt = µ + θεt−1 + εt h degrees of freedom AR(1) Yt = α + ϕYt−1 + εt H0 : The data is independently distributed (i.e. E(Yt ) = µ = residual correlations = 0, any observed corr. from α 1−β V (Yt ) = γ0 = σ2 1−β 2 , β =ϕ ARMA(1) Yt = α + ϕYt−1 + θεt−1 + εt Ps Seasonality Yt = β(t) + i=1 γi (Di,t )) + εt randomness of the sampling process) Ha : The data is not independently distributed, EWMA exhibit serial correlation 2 σt2 = 0 + (1 − λ)rt2 + λσt−1 Model parameters selection to Adj.R2 Covt = 0 + (1 − λ)rt,1 rt,2 + λCovt−1 Akaike’s information criteria (AIC) = 2k − 2 ln(L̂) 2 AIC = ln RSS n−k + n k GARCH Bayesian/Schwarz (BIC/SIC) = k ln(n) − 2 ln(L̂), 2 σt2 = ywvol + αrt2 + βσt−1 L̂ max value of the model’s likelihood function Covt = ywcov + αrt,1 rt,2 + βCovt−1 k – number of parameters y – pull towards long-run mean, reversion n – sample size BIC = ln RSS n−k + Long-run variance VL = k n w 1−α−β = w y Stable if α + β < 1 ln(n) SICk = nd/n = 50025/500 , d – degrees of freedom Augmented Dickey-Fuller (ADF) unit root test 2 R in trending timeseries → 1 H0 : unit root exists, in yYt−1 y = 0 Box–Jenkins model selection y = 0 means no predictive value from the past, e.g. 1. Stationarity via ACF PACF plots, ADF for unit random walk. roots, Jarque-Bera for N (0, 1) of returns Lagged level, deterministic, lagged differences; 2. Parameters estimation (AIC, BIC): OLS for AR, random walk with drift. max likelihood for AR, MA, ARMA ∆Yt = α + βt + yYt−1 + σ∆Yt−1 + .. + εt HA : y < 0 3. Residuals diagnostics (Ljung–Box for white noise) Conditional heterosk. is worse than unconditional. 10 Distribution Heterosk.: coefficients are unbiased and consistent, f (x) = P DF = CDF ′ st. ε are unreliable. CF D(x) = P (X < x), probability of outcome AR process is stationary when its lagged strictly less than x polynomial is invertible. PMF dice: f (x) = 61 , Treating non-stationarity Q(x) = N −1 (x) = F −1 (x), inverse CDF, quantile F (x) = x 6 Q(50%) – median R µ = xf (x) R P σ 2 = (x − µ)2 f (x)dx = (x − µ)2 P (x) Trend → estimating or removing Seasonality → dummies Unit root (spurious relationship, no mean reversion) → differencing Random Variables U = µ1 + σ1 Z1 V = µ2 + σ2 (ρZ1 + p 1 − ρ2 Z2 ) LLN: limn→∞ Ê[f (x)] = E[f (x)], µ̂ → µ 4 CLT: V (Ê[f (x)]) = σf2 /n, Ê[f (x)] ∼ N (µ, σf2 /n) m = ln µ and σ 2 are finite. s2 = ln(1 + w) [Se(µ−σ Mixture Y = wX1 + (1 − w)X2 2 2 E[Y ] = ρX12 + (1 − ρ)X22 Y = Y = 1 x 1 −β , βe V ar[x] = np(1 − p) P (x) = px (1 − p)n−x n n! x = x!(n−x)! Pv i=1 Zi2 V = 2v Y = √ Z2 χ /v E=0 V = Portfolio default rate v v−2 p 1 loan loss σ1 = (1 − RR) p(1 − p) p σp = σ1 n(1 + (n − 1)p) = p (1 − RR) np(1 − p)(1 + (n − 1)p) √ σ n(1+(n−1)p) σ αp = np = 1 n Power Law Beta Moments 1 α−1 (1 Beta(α,β) p K – scale α – fatness µ = E[X] − p)β−1 , mass ∈ [0, 1] V = E[(X − µ)2 ] E[(X−µ)3 ] σ3 E[(X−µ)4 ] = σ4 Loss frequency K (from exponential) K(X 2 Y 2 ) – parabola on corr/kurtosis plot λx e−λ x! K(XY 3 ) and K(X 3 Y ) – lines H0 : S = 0, K = 3 Uniform 2 JB = (T − 1)( Sk6 + (b−a)2 12 V = P (l < x < u) = (K−3)2 ) 24 vs χ2 Positive skew: mode, median, mean – mean is U2 = a + (b − a)U1 affected by outliers. min(u,b)−max(l,a) b−a 11 Normal E=µ v−2 v−4 Jarque-Bera N(0,1) test S= E= K= Fat tails P (v > x) = Kx−α , Poisson (discrete) a+b 2 V = 1/λ2 E = 1/λ Student P (X = x) = ] – confidence interval 1/β = λ V = β2 E=v Binomial (discrete) E = V = λ = np = σ µ Chi P (x) = px (1 − p)1−x t β 2 P D12 = SP1 − SP2 = e−h1 t1 − e−h2 t2 Bernoulli f (p) = ,w = Survivalto 6 = e−6/5 , β = 5 fx1 ,x2 (X1 ,X2 ) fx2 (X2 ) n x √ /2)t±zα σ t E=β Conditional E[x] = np µ (1+w) Exponential 2 V (Y ) = E[Y ] − E[Y ] fx1 |x2 = 2 √ VAR r = ln P1 − ln P0 , V = σ2 R= P1 P0 − 1, r = ln(1 + R) P1 P1 = P0 (1 + R) = P0 eln(1+R) = P0 eln(1+ P0 −1) P Q rperiod = r Rperiod = (1 + R) Lognormal Loss severity Coherent: Monotonicity E = µ + σ 2 /2 R1 ≥ R2 → p(R1 ) ≤ p(R2 ) p(R1 + R2 ) ≤ p(R1 ) + p(R2 ) Sample from Poisson → get n losses Subadditivity Sample n from N (0, 1) Pn 2 Loss = i=1 eN (m,s ) Positive homogenity Translation invariance 5 p(βR) = βp(R), β > 0 p(R + c) = p(R) − c VAR coherent when N(0, 1); ES always −z 2 /2 Scenarios −1.652 /2 e e √ √ ES = µ + σ (1−x) = µ + σ (1−0.95) 2π 2π DFAST 1/2y, no cap. plan, [10, 50b] Spectral: ↑loss ↑weight CCAR 1y, 9Q horizon, (50b, ∞) Xa = F −1 (1 − U ) Uantithetic = 1 − U Guassian Copula CBB – circular block bootstrap with replacement, √ block size = n √ Simulations: εa = ε 1 + ρ, ε = √σn √ V ARb = P V0 σy M Dzα t U1 = a1 F + p 1 − a21 Z1 Economy ↑ F ↑ Ui ↑ F0.1% = N −1 (0.001) W CDR(F ) = Conditional distribution has (µ, σ 2 ) conditioned on √ N [N −1 (P D)− ρN −1 (0.001)] √ 1−ρ EL = P D × LGD p 2 U L = σEL = P DσLGD + LGD2 σP2 D economic situation, it can be N(0, 1), but result in fat-tailed unconditional distribution. Control variable is effective when q 2 σ corr(x, y) > 0.5 σy2 Operational risk ∗ PA∗ = (PA − PBS ) + PBS SMA: 7x + 7y + 5z, Basic indicator (BI): 15% GI x y < 10m u = eσ Trees t e pu = of Gross Income (GI) −d u−d 12% retail, asset management rt 15% commercial banking, agency −d Sert = Su eu−d + Sd u−e u−d Futures pu = 18% corp. finance, trading, payment and settlement Approaches 1−d u−d fu −fd Su −Sd u −∆d Γ= ∆ S u −S d ud S u = S2u +S 2 ∆= 13 Basic indicator, firmwide, % of GI Sd = Standardized, business line ×β S2d +Sud 2 Advanced measurement, operational VAR AMA: Loss frequency = Sensitivities Loss severity µ µ̂ = ln √1+w DV 01 = D × 0.0001 × P ∆P = D∆S + 12 Γ(∆S)2 + V ∆σ + T ∆t mD = B− −B+ 2B0 ∆y z < 100m % (r−q)t rt mC = e−λ λn n! σˆ2 = ln(1 + w), Estimated lossy = Observed lossx B− +B+ −2B0 B0 ∆y 2 w = (σ/µ)2 0.23 Revenuey Revenuex Hedging D and C −V D0 − P1 D1 − P2 D2 = 0 15 V C0 + P1 C1 + P2 C2 = 0 Forward Trailing hedge: ρ σσFS F = S − I − Ke−rt = Se−qt − Ke−rt t 1+R , F = (S + U − I) (1+c)(1+q)(1+l) 14 1 − a22 Z2 Ui ≤ N −1 (P D) → def ault generating process in the underlying asset. √ p ρ = a1 a2 Contangion effect (σ ↑, ρ ↑) causes a different return 12 U2 = a2 F + S F Capital Financial Markets U - discounted storage costs Capital8% = CET 14.5 + AT 11.5 + T 22 h −1 i √ N (P D)+ ρN −1 (0.999) √ W CDR = N 1−ρ I - discounted coupons c - convenience yield Capitalirb = (W CDR − P D)LGD × EAD p 2 U L = P DσLGD + LGD2 P D(1 − P D) p U Lp = U L21 + U L22 + 2p12 U L1 U L2 q - dividend yield l - lease rate HR = β = a β Est. Lossa = Observedb ( Revenue Revenueb ) , β = 0.23 Cov(S,F ) 2 σF Hedge effectiveness 6 σ2 R2 = β 2 σF2 IR Collar f loor − cap CT D : CP − CF × F low Risk reversal, Range forward chigh otm − pitm S GP = CF × F + AI CF = GP −AI 100 , F Cliquet – portfolio of ATM forward starting options = 100 Asian (S − K)+ Rf wd = Rf ut − 12 σ 2 T1 T2 , σ – 1y volatility of future rate Lookback T2 = T1 + 90d Fixed: c = (Smax − K)+ Rf ut = 100 − Zquoted Floating: c = (Smax − ST )+ p = (K − Smin )+ Eurodollar Future Gap has trigger K0 , payoff K1 , possible neg. = 10000[100 − 0.25(100 − P %)] = 1m(0.75 + 0.25P ) premium Compound – option on option FRA Shout c = max[(S − K)+ on tshout , (S − K)+ on T ] FRA 1×4 – 3m forward rate in 1m Rainbow – option on diff. assets F −RK )τ −rt F RA = L (R1+R e Fτ Volatility swap L(σ − σK ) Option 2 Variance swap L(σ 2 − σK ) E(St ) = S0 eµt Warrant price delusion E(lnSt ) = lnS0 + (µ − Stock value decline S − σ2 2 )t n n+m c m n+m c V (lnSt ) = σ 2 t Longevity L(Kmortality − R) S − K ≤ C − P ≤ S − Ke−rt c p C P Chooser max(c, p) = c + max(0, c − p) = c + (Ke−rt − S)+ = p + (S − Ke−rt )+ S + - + - ∆call − ∆put = 1 X - + - + ∆call = e−qt N (d1 ) T ? ? + + ∆f wd = e−qt σ + + + + vanna ∆ to σ r + - + - charm ∆ to time Div - + - + vomma vega to σ √i 2 lnSt ∼ N lnS0 + (µ − σ2 )t, σ t √ )t−Zσ t σ2 2 < St < elnS0 +(µ− √ µ̂ = µ − σ 2 /2 σ̂ = σ/ t ∆f ut = e(r−q)t ′ σ2 2 ′ (d2 ) √ gamma = e−qt NSσ(d√1t) = Ke−rt N S2 σ t √ √ vega = Se−qt tN ′ (d1 ) = Ke−rt tN ′ (d2 ) h elnS0 +(µ− ∆put = e−qt [N (d1 ) − 1] √ )t+Zσ t rhoc = Kte−rt N (d2 ) rhop = −Kte−rt N (−d2 ) c = S −qt N (d1 ) − Ke−rt N (d2 ) C is exercised early when Dt > K(1 − e−r(T −t)) p = Ke−rt [1 − N (d2 )] − S −qt [1 − N (d1 )] P is exercised early when Dt < K(1 − e−r(T −t)) d1 = −qt ln( S K 2 )+(r+ σ2 √ Asset or nothing put pays 1 unit of asset, when S < )t σ t √ d2 = d1 − σ t K (Z graph). Fiduciary call c + Ke−rt = p + S Protective put On stocks – American, on index – European. Covered call S − c Dividend in stock ≈ stock split → exchange adjusts Box spread r strike. Straddle c + p Rho is highest for ITM. Strip 2p + c Theta is highest for expiring ATM. Strap p + 2c Butterfly, calendar spread citm − 2catm + cotm Diagonal = calendar with diff. K CBOE Margin high high low Collar p + S − c ≈ clow itm − cotm = potm − pitm Bull −c : max(c + 0.2S + (S − K)− otm , c + 0.1S) call & put spreads (no S) −p : max(p + 0.2S + (K − S)− otm , p + 0.1K) 7 Forward KR010−1Y = Bond c2 2 + .. (1+ R1 +0.0001 ) 2 KR01 by 1bp, Dk r by 100bp 1+ MM and T-Bill Act/360 T-Bond Act/Act Corporate r e = (1 + y/m)m c1 R1 +0.0001 2 Yield curve: Bull flattener 30/360 Bull steepener n n = 1 − 0.068 360 P T Bill = 1 − Rq 360 Rq = (1 − P ) 360 n Terms: c Par yield: 1 = A m +d= 2(1−DT ) V = AT 1+Rnom Rreal = 1+Rinf PT = 1+ P c DFi m +d c−PT 2 1+ + short↓ long end↓↓ short↓↓ long end↓ MM 1Y Short 1-5 Medium 5-12 Long 12+ E(Rbond ) = RT reasury + SDP − P D(1 − RR) AT Defaulted Zcpn claim receives issuing price +AI. Mortgage T-Bill rates go lower than OIS on FED fund rate CP R = 1 − (1 − SM M )12 due to capital requirements. 1 SM M = 1 − (1 − CP R) 12 Pn D = P1 i=1 ti ci e−yti M D = Pn C = P1 i=1 t2i ci e−yti M C = D 1+y/m PSA ← calculator 0.2% CPR to 30m increasing by 0.2% every year; C (1+y/m)2 6% flat afterwards. 1 a+be−cI P repayment R = PAC – planned amortization mortgage tranche I = (W AC − R)ALS × A − K, IO lower in value than PO, IO can receive less than WAC – weighted avg. coupon paid due to prepayment uncertainty. A – annuity factor Cash-out refinancing – extracting home equity. K – cost to refinance h i 1 A= m 1 − mT y (1+y/m) h i 1 c m V alueA = m y 1 − (1+y/m)mT discounted using T-rate + OASestimate, until PV = MTM. c y V alueP erpetuity = Japan bonds y = OAS is computed via MC simulation where CFs are c p + 2 MBS with equal credit quality, buy one with 100−p pT bigger OAS. Cash & Carry = N ct − P0 rt Rrealized = OAS = Zero volatility spread – option cost P1 −P0 +N ct−P0 rt P0 P1 +Cash & Carry −P0 = Carry-roll-down +Rate Change +Spread Change Dollar roll -Aug TBA +Sept TBA Dollar roll value P0 − P1 − N ct + P0 rt Prepayments↑: PO↑, IO↓ Corp Bond 105-07 = 105 78 7 T-Bond 105-07 = 105 32 18 6 Futures 105-187 = 105 32 256 = 105.5859375 1 1 256 V = P 2 3 5 6 2 256 3 256 4 256 5 256 σi2 Factor 7 8 6 7 256 256 2 σ importance= Vi KR01 = P V1 − P V0 Dkr = P V1 −P V0 P V0 ∆ykr Dkr = 10000×KR01 P0 = KR01 P V0 ∆ykr D= P Dkr 8 16 SONIA, ESTR, EONIA are interbank rates Timeline Dutch auction sells at the first full offer P 1929 Wall Street crash Yankee bonds – bonds by international 1933- Glass–Steagall (GS) segregated inv. organizations banking Make-whole call provision has no cost to 1970 inflation↑, FED short term rate↑ bondholders 1988 B1, capital 8% Unsystematic risk is eliminated by 30 stocks and 1990 Brazil on loc. cur. debt thousands of bonds 1993 Oil↓ $15/barrel Pass-through mortgages carry prepayment risk 1997 Asian crisis, S&P -7% TBA – to be announced, forward mortgage market 1998 Russia on loc. cur. debt, flight to safety, Through the cycle PD: growth – overstated, crisis – Oil↓ understated -1999 GS repeal RCSA – risk control and self assessment 2001 9/11, Enron Knock-on effect – respose exacerbating adverse 2002 World Com, Global Crossing condition 2003 Parmalat, Sarbanes–Oxley (SOX) Asymptotically normal means becomes normal as 2007 US subprime, housing n→∞ 2008 Lehman counterparties, Bear Sterns, Girsanov theorem σrn = σreal Merrill Lynch, Libor-OIS spread 3.5%, Spurious regression – false relationship Northern Rock, Oil↓ Contraction risk – prepayment risk 2009 G20 limits bankers’ bonuses BCBS operational risk types 2010 EU sovereign debt, Dodd-Frank, Volcher Internal, ext. fraud, employee practises, and rule: no prop. trading or fund ownership workplace safety; ×2 Greece defaults on for. cur. debt, EU Clients, products, and business practises; bans uncovered CDS, LIBOR Damage to physical assets; manipulation scandal Business disruption and system failures; 2012 2014 Oil↓, SEC rules that originators must retain 5% of securitized product Execution, delivery, and process management. KMV model is PIT in contrast to agencies rating 2016 UK leaves EU, FRTB models. 2020 Oil↓ Euler’s theorem to compute individual loan contribution: 17 ∆F ∆Xi /Xi , Terminology ∆F = p (X1 + ∆X1 )2 + X22 + .. Orders World’s theorem: every time series can be written market as a sum of deterministic and stochastic time series. limit Yt = εt + b1 εt−1 + b2 εt−2 + .. + ηt , stop-loss, sells below MV ηt – deterministic sin process ∼ market if touched, sells above MV Jensen’s inequality: convex → E[f ] > f (E) stop-limit PDF gives probability density of X, can be above 1 good till cancelled (open) Control variables – var. with known relation to Y fill or kill VIX – 30d implied volatility of SP500. Extraneous variables – superfluous var. with β = 0 SOFR repo-based secured ON financing rate 9 Positive definite – every linear combination of Xi PCAOB – Public Companies Accounting Oversight must have a non-negative variance. Board (from SOX). Matrix corrections NINJA loan – no income, no job, no assets. Equicorrel. – all correlations are equal Liar loan – no evidence of employment. TAF – term auction facility 2008 PDCF – primary dealer credit facility Troubled assets relief program Correlations are to the same factor ρij = Yi Yj Uncovered interest rates parity – rates are the same, but one currency depreciates. IR↑ Bond↓ Forward↓ Futures↓↓, as loss has to be financed at higher rate. GARP Knightian uncertainty – known unknowns Conduct and integrity RMP Principles Conflict of interest 1. Identify Confidentiality 2. Measure, manage, map 3. Operationalize risk app., distinguish EL and UL Responsibility Professional standards Best practices 4. Address the relationships 5. Implement plan or strategy 6. Monitor and adjust 18 Insurance Senior sets, BU implements, Fin. and operations P D12 = SP1 − SP2 = (P D12 |SP1 )SP1 mitigate and transfer, RM supervises. (P D23 |SP1 ) = RM is more concerned with UL. (P D12 |SP1 ))(P D23 |SP2 ) = 1 − LGD↑ bankruptcy (liquidation) risk Find Premium Actual risks < Appetite < Capacity 1) P D23 |S12 = P D23 (1 − P D12 ) Concentration limits do not counter correlation risk 2) P V = N P D12 (1 + R −1 2) Financial position risk – balance sheet risk 3) n = 1 + 1 × S12 (1 + R −2 2) Dodd-Frank 4) P remium1Y = P V /n 1. FED oversights SIFIs (>50b) Ratios 2. Ends to big to fail Loss = Paid / Premium 3. Living will Expenses = Expenses / Premium 4. Derivatives markets Combined = (Paid + Expenses) / Premium 5. The Volker rule Combined after dividends = (Paid + Expenses + 6. Consumer Financial Bureau Dividends) / Premium 7. DFAST (10), CCAR (50) Operating = (Paid + Expenses + Dividends - CDO of mortgages is CMO. Investment Income) / Premium CLO of bank loans default less than mortgages due Against moral hazard SP2 −SP3 SP1 = P D23 SP 1 = (1 − SP1 −SP2 SP1 P D23 SP2 + P D23 |S12 (1 + R −3 2) Deductibles to better credit process. Coinsurance provision Reassignment – transfer risk to 3rd party in the Policy limits Mortality risk is bad for life insurance, good for event of downgrade. SIV – structured investment vehicle to profit from annuity business. spreads. Mortality risk is mitigated by shorting longevity Contango – no benefit of holding the asset. derivatives (fixed - actual mortality) and survival RDARR – risk data aggregation and risk reporting. bonds (coupon is linked to the number of survivors). 10 Minimum CR = 0.25 to 0.45 × Solvency capital 19 requirement Derivatives & Integrals Below SCR → plan to increase (x)′ = 1 R Below MCR → business operations restricted (af )′ = af ′ R 1 = x [+C] R af = a f Plans (ax)′ = a R a = ax Defined benefit plan (employee benefit fixed) (f ± g)′ = f ′ ± g ′ R f ±g = Defined contribution plan (employee benefit (f g)′ = f ′ g + f g ′ unknown) ( fg )′ = f ′ g−f g ′ g2 a Rb (f (g))′ = f ′ (g)g ′ Life insurance Whole life on death, fixed premium (f n )′ = nf n−1 f ′ Term death in period, fixed premium (xn )′ = nxn−1 Endowment at the end or on death (ln f )′ = f′ f opposite, stops on death (ln x)′ = 1 x = x−1 (loga x) = x ′ ( ln ln a ) Annuity Rb f± R c = c(b − a) xn = xn+1 n+1 R 1 ax+b = R ln x = x ln x − x R loga x = (ef )′ = ef f ′ R ef = (eax )′ = aeax R eax = a1 eax (bax )′ = a ln(b)bax R bax = = 1 x ln a g f = F (b) − F (a) R ′ 11 a R 1 a ln(ax + b) x ln x−x ln a 1 f f′ e 1 ax a ln b b