Quantitative Analisys

Linear Regression Formulas Statistical test Volatiltiy and Value at Risk Important info
√
β= Cov[x,y]
= ρ σσxy White Test: It is a test of eteroschedicity. The null Condition for T rule Condition to apply the rule
var[X]
hypothesis is the linear regression of all the explana- for volatility:
βBiased = β1 + β2 ρ σσ12
tory variable and cross products of them against the
p
S.E.(β) = √nσ s σt = t/T σT
x errors ϵi must contains only parameters not significa- is that risk factor timeseries be iid. Otherwise, if ti-
double tail p value = 2(1 − ϕ(T )) tivley different from zero:
R2 1 par = ESS RSS meseries is correlated, we need to apply: σrt,t+2 =
T SS = 1 − T SS ϵi = γ0 + γ1 X1 + γ12 X1X2 + γ2 2X 2 + γ2 X2 σrt,t+1 + σrt+1,t+2 + 2Cov[rt,t+1 , rt+1,t+2 ]
R2 k par, adjusted = 1 − RSS/(n−k−1) F-test on γs insignificant.
T SS/(n−1) that is a greater value.
F-Test Test for the linear regression of multiple ex- Ghosting Effect is the effect that is created for ins-
planatory variables. The idea is to get the R2 of a tace by MA. When a past observation disappear from
TimeSeries analysis Formulas unrestricted model (all explanatory variables) and res- window you are estimaating volatiltiy on, there can
AR(1) : tricted model (model without the variables you want be a great change in today volatility justified only by
rt = δ + ϕrt−1 + ϵt to test the explanatory power). The test statistic is a a past obeservation disappearing.
ϵt = (1 − ϕL)rt F-statistic2 with
2
value: Approaches The approches for estimating volatiltiy
(R −RR )/q
M A(1) : F = (1−R2U )/(n−k u −1 can be Parametric and Non-Parametric. Under pa-
U
rt = δ + ϵt + θϵt−1 VIF : test for multicollinearity between explanatory rametric apprach we have: GARCH, EWMA. Under
rt = (1 + θL)ϵt variables. You regress a variable on all the others and non parametric we have Historical Simulation, Multi-
ARM A(1, 1) : then compute the R2 . Vif is given as: variate density estimation.
1
rt = δ + ϕrt−1 + θϵt−1 + ϵt V IF = 1−R 2 Var properties Var is not a coherent function, like
If VIF is greater than 10 then variable is multicolli- volatiliy, expected shortfall or other. A measure to be
near. choerent must be:
Measuring Volatility and Value at Risk formulas Cook’s distance it test the presence of outliers. You
need to create models without point you want to test 1. Subadditivity: V AR[A, B] ≤ V AR[A] +
GARCH(1, 1) :
as outlier and the model with all point. The distance V AR[B] where A and B are two portfolios (or
σt2 = ω + αrt−1
2 2
+ βσt−1
ω between thisPtwoj models is defined as : single assets). VAR doesn t decrease always on
LongRunAverage = 1−α−β
CooksD =
(Ŷi −Ŷi ) diversification.
σn+t = LongRunAverage + (α + β)t (σn − ks2
LongRunAverage) where k is the number of explanatory variable and s 2. Monotonicity: If expected value of A is greater
EW M A : is the variance (from MSE). If ¿ 1 possible outlier than B, than risk of A is less than B. So we
σt2 = ασt−1
2 2
+ (1 − α)rt−1 α = 0.97 Liung Box Q: Test for autocovariance of a timese- shoud expect V AR[A] ≤ V AR[B]
∆t
q
∆t
ries.
V ARp/i = −µp T + T σp zα
Pm
QLB = T (T + 2) τ ρ2 (τ )( T −τ 1
) where τ is lag, ρ is 3. Translation Invariance: If you add cash reduce
pdfα σ
ESα = 1−α
autocorrelation function, T the sample size risk: V AR[A + Cash] < V AR[A]
Box Pierce Q: Test for autocovariance of a timese-
ries. 4. Positive Homogeneity: greater size of portiolio
TimeSeries analysis Important info Pm
QBP = T τ ρ2 (τ ) means greater risk: V AR(λA) = λV AR(A)
Stationarity conditions: for AR The root of the
where ρ is autocorrelation function, T the sample size. VAR lack subadditivity.
characteristic polynomials of a AR must be outside
BoxPierce and Liung Box have a χ21−α,m statistic,
unit circle (other wise there is a unit root).
where m is the number √ of lag you are testing, α is
For MA, they are always stationary. But they can be Distributions
treshold p-value. m ≃ T
not invertible if the prevoius condition is not realized. 2
σBinomial = nP (1 − P )
Dickey Fuller/ Augmented Dickey Fuller: Test
for unit root. σχ2 = 2ν
Probability σP2 oisson = λ
2
Bayes: P (A|B) = P (B|A)P (A)
. Or: P (A ∪ B) = σstudent = ν/(ν − 2)
P (B)
P (B) + P (A) − P (A ∩ B). And: P (A ∩ B) =
P (B)P (A) = P (B|A)P (B).
 Valuation and Risk Models
Binomial trees Trading Strategies Exotic Options
ert −d Covered Call: S − C Gap there is a trigger level K2 and a strike K1. when
risk neutral probabiltiy: p = u−d
if you know σ: u = eσt ; d = e −σt Protective Put: P + S pricing you need to use k2 in computing d1 and d2 of
(r−d)t Principle Protected Note: P V (K) + BS.
stock with dividend: p = e u−d−d
C(atT heM oney) Forward start Today is T, option start at T1.p =
e(rdomestic −rf oreign )t −d
foreign exchange option: p = u−d Bull Spread : K2 > k1 ce−qT 1 q dividend yield.
1−d
Futures options: p = u−d Buy lower strike, sell greater Cliquet Forward start options in which strike is de-
1−d Bear Spread : K2 > k1 cided in future.
Futures options: p = u−d
−fd
∆ = fuu−d Buy greater strike, sell lower Compund Option on Option.
Box Spread : Bull + Bear Chooser at maturity, you can choose the type of op-
it worth K2-K1, and price PV(K2-K1) tion (put or call).
Black-Scholes Model Butterfly: k1 < k3 < k2, K3 in the middle Binary They can be cash or nothing (PV(Q)N(d2))
Non dividend pay stock: Buy K1, Buy K3, sell 2 K2 or asset or nothing (PV(S)N(d1)) . Sum of cash or
C = SN (d1) − P V (K)N (d2) Straddle: Call and put on Same strike. nothing and asset or nothing european call.
2
ln( S
k )+(r+
σ
2 )T Write Straddle when you sell both, purchase straddle Barrier The payoff depend whether the strike touch
d1 = √
σ T
σ2
when you buy both. or not the barries during life. There are knock in and
ln( S
k )+(r− 2 )T
d2 = √
σ T
strip on the same K, (Purchase) you are long on 1 Knock out. there is the rule that a knock out plus the
call and long on 2 put associate knock in is a european option.
dividend pay stock: strap on the same K, (Purchase) you are long on 2 Parisian It is a barrier but the option must be above
C = Se−dT N (d1) − P V (K)N (d2) call and long on 1 put the barrier for a certain period.
2
ln( S
k )+(r−d+
σ
2 )T strangle Same as straddle but with different K. So Lookback The payoff depend on the history of the
d1 = √
σ T
σ2
K1 < K2, buy put on K1 and call on k2 stock: floting call: S-Smin; floating put: Smax-S; fi-
ln( S
k )+(r−d− 2 )T
d2 = √
σ T
xed call Smax-K; fixed put: K-Smin
Vanilla Options Asian The payoff depend on average of the price his-
FX option: tory. it can be: Average price: Saverage- K; Average
Put - Call (european) Parity: P + S = C +
C = Se−rf oreign T N (d1) − P V (K)N (d2) strike; S-Saverage. usefull for fx exposure.
σ2
P V (K)
ln( S
k )+(r−r√
f oreign + 2)T Volatility Swap There is a swap between the expec-
d1 = σ T
Put - Call (American) Parity: S − K ≤ C − P ≤
ted volatility and realized volatility on index/stock.
ln( S
k )+(r−rf
2
− σ )T S − P V (K)
d2 = √oreign 2
σ T bound (european) call S ≥ C ≥ S − P V (K)
bound (european) Put P V (k) ≥ P ≥ P V (K) − S Credit Risk Formulas
Futures (Black-62): bound (American) call S ≥ C ≥ S − P V (K) EL = P D ∗ p
LGD = P D ∗ EAD ∗ (1 − RR)
C = F N (d1) − P V (K)N (d2) bound (American) Put k ≥ P ≥ K − S 2
U L = EAD P D ∗ σLGD + LGD2 ∗ σP2 D
2
ln( F
k )+(
σ
)T 2
d1 = √ 2 In case of dividend pay stock you need to subtract Remember σP D = P D(1 − P D)P
σ T U Li ( U Lj ρj )
ln( F σ2 D (present value of Dividend) in the lower bound Risk Contribution U LCi = alll oansa lsoi
k )+(− )T
d2 = √ 2
σ T
U Lp
Risk weighted asset ECi = U LCi CMi √
American Call option should be exercised only on σ 1+(n−1)ρ
Dividend pay stock when D = K-PV(K). Standard deviation of credit losses α = √nEAD
Greeks p √
An American Put can be often be exercised early. one factor model Ui = ai F + 1.a2i Zi with ρ = a2
∆ = N (d1) √
N −1 (P D)+ ρN −1 (x)
′ Vasicek W CDR(X) = N ( √ )
Γ= N (d1)
√ 1−ρ
Sσ T √
V ega = Sσ T N ′ (d1)
Black-scholes-merton PDE can be seen as
Θ + rs∆ + 12 σ 2 S 2 Γ = RΠP
Portfolio Delta: ∆p = wi ∆i
 Credit Risk info Interest Rate Risk Pricing and conventions
Risk Capital: The idea is that a bank need to Macaulay Duration Average time to maturity Annuity: Y 1/k [1 − (1+y/k)1
kt ]
structure its capital and the risk capital can be seen Modified Duration M acaulayDuration m payment
1+y/m Treasury Bill: QUoted Price = 360 n (100 −
as all its capital at risk , that it take from sharehol- frequency
P −P CashP rice), n is remaining life. This means that the
ders, creditors, depositors. If equity fall down the Effective Duration y−ϵP0 2ϵy+ϵ quoted price is a rate, but it is not the real yield that
liabiltiies, the critical treshold for a bank to be still
Py−ϵ +Py+ϵ −2P0 is:
considered as still ”going concern”is the limit between Effective Convexity 100Q
P0 (2ϵ=2 yield = Cash
junior debt and senior debt. If this is exceed, this can
DM odif ied ×P rice Bonds in America are 30/360, corporate bond are of-
lead a downgrade or a ”gone concern”. The economic Dollar Value 01 DV 01 = =
10000 ten Act/act.
capital is the buffer banks put aside to avoid this . Dollarduration
10000 Treasury Bond Future: A treasury bond future
Economic Capital: It is generally defined as:
allow the seller to deliver the chepest to delivery bond
EC = U L − EL Duration × ∆Y = P ercentageChangeInP rice (CTD). To get it you need to minimize the difference
under IRB, this must be estimated as : DV 01 × ∆YBP S = P ercentageChangeInP rice between cash received by seller (that depend on con-
(W CDR − P D)LGD ∗ EAD Properties: tract feature) and cash prices of bonds:
Capital Multiplier Under the standardized appro-
CostOfDelivery = QUotedPrice + AccuredInterest
ach of Basel II the idea is to attach a multiplier to the 1. Increasing Coupon, Duration decrease
CashReceived = SettlmenentPrice X ConversionFac-
risk contribution of an instrument in order to get its
2. Increase coupon, DV01 increase ( Pull to Par in tor + AccruedInterest
capital requirement (regulatory capital):
reverse) The CTD is the bond for witch Cost - Cash is mini-
ECi = U LCi ∗ CMi
mum
capital multiplier represent a percentile. 3. Fixing coupon and maturity, increasing yield, For the pricing of a Future you need to account:
duration increase F utureP riceOf Bond = (S − I)erT − AI, where S is
Mortgage/Mortgage Back Securities the cash price today, I is the PV of cash flow that will
4. Becasue Portfolio duration is the weighted sum
AmountBorrowed P = be received from now to maturiy, AI is the accreud
12T 1 of instruments duration you can create portfo-
M onthlyP ayment i=0 (1+y/12)n interest in time to delivery. When you have this:
lios with a certain duration with two strategies:
AmountBorrowed = M onthlyP ayment 1
[1 − (1+y/12)12T ] F utures = FConversionF
utureP riceOf Bond
actor
y/12
1) Barbell: Long duration and low duration to
In the calculator you can simply compute this amount get a median portfolio duration. NB, must be
Modern Portfolio Theory, APT and performance
creating a bond with 0 FV and PV equal to the computed weighted on PRICE not Fave Value
Capital Market Line CM L : E[Rp ] = r +
amount borrowed. Moving then on Bond computati- 2) Bullet: all bond with the desired durations. E[Rm ]−r
ons you can find also outstanding and principal pay- σm σp
ments. 5. Notice that effect of convexity ( that is positive Sharpe Index SP I = E[Ri ]−r
σi
Mortgage Backed securities are pool of bonds, prin- in Bond) is that a loss computed from dura- E[Ri ]−r
Treynor T P I = .
Should be equal to
cipal and interest cash flow of it are sell back to in- tion will be greater in reality and gain computed βi
from duration will be lower in reality. E[Rm ] − r at equilibtium.
vestors thanks to an SPV and a mortgage guarantors Jensen’s alpha It is the alpha of a CAPM like
that pay the principal in case of default of a bond in regression:Ri − r = α̂ + β̂i (Rm − R). Should be 0
the pool. So the main risk the investors are exposed Appendix
at equilibrium
is the prepayment risk. Euler’s theorem: Given F homogeneous risk mea- Sortino ratio More general focus on the down-
Fro MBS we have: sure P
(as VAR) we can decompose it as : side risk. Given a benchmark return r: SR =
WAC: weighted average Coupon F = i Qi √P Rp −r
∆F 2
min(0,Rpt −T )
WAM: weighted average maturity Where Q1 = ∆X i
i.e. the risk measure computed only history

SMM: Single mortality rate X1 Information ratio AverageActiveReturn

T rackingError
on one risk factor divided by percentage change of the APT it implements factor analyis, so under this um-
CPR: Conditional prepayment rate
change in risk factor. brella there are Fama-French model and CAPM. It is
CP Rn = 1 − (1 − SM M )12
There are way of modelling refinancing and prepay- used to construct 0-beta portfolios
ment rate.
I = (W AC−R)AverageLoanSize×AnnuityF actor−
CostOf Ref inancing
CP R(i) = T + a+be1 −bI
 Financial Markets and Products
Pricing Forwards and Futures Insurance Funds
General idea for pricing of forward ( financial or con- Term life/life Insurance Usually you have a table Mutual funds They have usually lower fees, due to
sumption investment) it the idea of cost of carry. with following headers: age;Conditional Death Pro- disclose investment strategies, and subject to regula-
The cost of carry is like to see the time you have an bability; Cumulative Survival. tion they are divided in 3 categories:
asset as time in which ther are sure expenses. There To compute premium:
1. Open End: they disclose the NAV each day at
is the expenses of time, this is why the risk free rate
1. Compute Expetected payout: actually it is No- 4pm, the outstanding share change almost daily
always enter in computations, there is the convenience
minal X ( PV(probability of death year1) + (possibility to liquidate).
yield, that is the income that are related to carry the
PV(probability of death year 2) ...)
asset (for instance lease rate in commodity case) and 2. Closed End: Since they doesn’t disclose daily
starage cost. Generally, total cost of carry is : 2. remember: Prob Deat year2 = CondProbDe- the NAV and there is no possibility to liqui-
riskF ree + storageCost − ConvenienceY ields atYear2 X (Survival Year1), where survival = date, the outstanding share is fixed, and there
So a forward price must accout of all of this In fi- 1-PorbabilityDeatYear1 is a Back-End fee (to exit position). For this
nancial assets case, we consider storage cost equalt to reason, actually the price of the share is below
0. 3. The PV for the payout is discounted at half year. NAV.
General Forward/Future: F = Ser+sc−cy . (so first year in 6 month, 1 year in 1.5 year and
So so on) 3. Exchange-traded: this are fund that have shares
Equity Forward/Future: F = Se(r−d)T . traded on exchange. nav and price is almost the
4. After computing Expected payout, to com-
Fx: F = Se(r−rf oreign )T . same since they have great liquidity (provided
pute premium you need to solve the fol-
Commodity: F = Ser+sc−cy . In the commodity by financial institutions).
lowing equation: P remiun + P V ((1 −
case this formula is really clear. For the theory of
DeatY ear1)P remium)... = ExpectedP ayout Hedge funds: high fees. fee applied is 2 and 20,
expected future price, a price of a commodity future
where 2 is management fee and 20 is performance fee.
should be Whole life coverage for all the life Variable life Performance fee has some constrains: usually it is de-
F (0, 1) = E(S(1))e(r−k)T where k is the discount rate part of the surplus is invested Endowment life The fined an hurdle rate, that is the return that must be
of the commodity future, Because of this , future curve premium is paid even if you not died. Group life exceeded to trigger performance fee. When the hurdle
can have swing: group of peaple with a single policy. Annuity pen- rate is exceeded once, then it is defined the so called
1)contango : when F¿E(S), convenience yield is low 2) sions schemes. Property casualty they have the ”high water mark”that is another higher return than
backwardation: when F¡E(S), when convenience yield most funds Health some measures: loss ratio to- must be exceeded again to trigger performance fee.
is higher than storage. talPayout /totalPremiums Expense ration Expense To protect investors there are clawbacks, from which
Excluding Payout / Premiums Combined ration investor can claim some past fees.
Foreign Exchange Market Loss ratio+ Expense ratio CRAD CombinedRatio Hedge fund has different possible strategies:
(1+r
yyy + Dividend/Premium Operating ratio Loss + Ex- Long/short equity: it is the idea to acquire a posi-
Interest Rate Parity: Fxxxyyy = 1+rxxx S
pense + Dividden - Invest Income Operatin ration tion to equities to profit the misprice. A portfolio can
Power Purchase Parity: Inf lationRateyyy −
must be less than 100% for a protibale business Pen-
Inf lationRatexxx = ∆S
S
be settle in different way to have a equity-market neu-
sion Plan At the end of the contribution you have tral, Beta Neutral, dollar neutral, sector neutral port-
the following payment: folios Dedicated Short: a portfolio with short po-
Banks X70%SalaryOf LastT OtY ear × sitions to selected companies that are seen as overva-
some assesement like: Y earOF contribution lued Distressed securities: Buy junk bonds , CCC
Net Stable funding Ratio: N SF R = To compute contribution rate (how much you need bonds Merger Arbitrage: when 2 companies mer-
availableStableF unding
RequiredStableF unding to pay monthly during work period): ges, there are some arbitrage opportunities that can
NSFT estimate liabilities Y earOf Contribition × ContributionRate × be exploited. Bet on the results. Convertbile arbi-
Liquidity Ratio: Current = Asset/Liabilities X = X70%SalaryOf LastT OtY ear × trage: Buy convertible Bond Fixed Income Arbi-
Quick = (Asset − inventory)/Liabilities Y earOF contribution trage Emergin Markets Global Macro macroeco-
estimate assets nomic analyis for the portfolios Managed Futures:
Leverage = debt/equity this is the most ”tecnical”usually they used tecnical
Cost of equity capital: Return a company need analyisis of fundamental analyis for this.
to bring to investor (ROI)
 Foundations of Risk Management
Learning from financial disaster Anatomy of great financial crisis Stress testing priciples
SavingLoans This companies ”rided”the yield curve, Only some points: Only some points:
borrow at short term and loaning long term. A flat- There are some principles to follow for stress testing:
tening of curve create default. Interest rate risk. 1. House price decline: the business model of
Lehman Brother, Northern Rock, Illinois Bu- majority of bank is US were the originate-to- 1. Should be done regularly, especially in period of
siness Model of Leman Brother was to sell loans and distribute, in which loans were sell back to in- expansion
then sell the cash flows to investors. This creates a vestor thank to securitization scheme. There
were the so called subprime market, in which 2. Stress events probabilty is difficult to estimate
leverage ratio of 31/1. Similar case in northern rock (not like Var)
that have a lot of loans that went to default. Liquidity loan were distribute to peaple with really low
risk credit worthyness (NINJA loan, no income, no 3. GFC rely on stress test that was only based
Metallgesellshaft this company used a dynamic job, no asset). Because of this subprime mort- on historical data with insufficiently aggregated
strategy hedging long term forward on oil (fixed), with gages and hosue decline, a lot of us citizen start risk measures.
short term futures. Then market shift to contango to default. This bring to a system collapse.
and roll down strategies was costly. (oil price fell). 4. Managemetn shoud be expert and need to con-
2. Liquidity problem, Short term funding: another duct stress testing (on details)
Hedging risk Long term capital management they point of business model was to have short term
had a strategies based on the assumption that market debt to finance long term assets (mainly with 5. the board must approve the risk appetite sta-
in one day cannot drop for more the 5%. When the shadow banking). This bring at great system tement , and management must implement also
market drop this caused hige losses. Model risk. vulnerability. performing stress test on business.
Baring the trader was allowed to get really specula-
tive position on Nikkey 250. This lead a great loss. 3. intermediaries: problem in the risk aggreagation 6. Documentation and good IT system are impor-
Rogue trading and reporting tant
Orange County Complex Derivative risk. One
should understand also the underlying risk associated Response to crisis 7. AUdit must be capable to understand stress test
to complex derivatives. and the capabilites of management
Only some points:
Volkswagen reputational risk: they modified car to
pass controls, but in the street they wasn t compliant 1. Sarbanes Oxley: it is the response to enron go- Risk Management
Enron Governance risk: This contains a series of vernance crisis. It rule the executuves to dis- Only some points:
fraudolent decision took from management, for ins- close their financial situations.
tance, when electricity drop in california, they deci- 1. in the ERP, the risk is no more interprete in
ded deliberatly to close power plant during demand 2. Dodd Frank Act: US reponse to financial cri- ”silos”, menaing centralaized and not divided
peak for raise price. And then they also cover some fi- sis. Some main point are: end too big to fail, in buisiness lines.
nancial statemnet. The Enron case bring to Sarbanes strenghten the fed, give stress testing to be con-
ducted regularly by bank like DFAST (bank be- 2. the CRO have the due to implement risk appe-
Oxley s act.
low 10b) and CCAR (bank above 10b) tite on the business line
SWIFT Cyber risk, there were an hacking of money
transactions. 3. SREP: european response to the crisis, it also 3. To implmenent risk measure it is important to
rule bank to conduct stress testing , called conduct regularly stress testing and scenario
EBA stress test, and reporting like ICAAP and analysis.
ILAAP.
4. BCBS 239: principles for stress reporting, some
main points are: P.3 Accuracy and integrity; P.4
Completeness; P.5 Timilinees ; P.6 Adaptability
 Something more
Carry-roll down more on loans/interest rate
Idea is to price bond in the future. There are two real rate/nominal Rreal = 1+rnominal
−1
1+rinf l
main approach: realized forwards, and forwards un-
changed. in the interest rate parity the rate are the nominals.
For the realized forwards approach you need to price
the bond in the next periods considering that: Loans rate Interest rate = Expected Loss - Opera-
(1 + r) = (1 + f1 )(1 + f 2)... ting Ratio - Margin
So multiplication of all the forwards within the period. Insurance operating ratio = loss ratio + expense
so if r is 1 year, the two forward you will multiply will ratio + dividends investment
be 6-12;12-18 (the 0-6 will be passed in 6 month). The DV01 for a portfolio is simply the sum of the
In the forwards unchanged you’ll need to consider the DV01s of the components of the portfolio
forwards THE SAME in the next period in both value
and reference period. So you will use for instnace if
r is 1 year, the two forward you will multiply will be
0-6;6-12
NB. the formula for the carry roll down is:
CarryRollDown = nextP rice − P reviousP rice +
CashCarry

Non parallel curve shifting

Key Rate 01 The main formula to remeber is that
a shift in a different bucket can be computed as (for
instance if key rate is 5 years and 2 years and you
want to compute shift in 4 year):
Shf it4Y ear = 4−2
5−2
Forward bucket 01 You need to use the formula:
(1 + r) = (1 + f1 )(1 + f 2)...
to shift only certain f and find back the spot rate r.

Commodity Forwards Shape

Agricultural High seasonality and varying storage
cost. Forwards curve has swing.
Metal not seasonal and low storage cost. Anyway,
lease rate affect a lot the forward curve as they are
used as investement asset (not holding phiysically)
i.e. gold.
Energy Crude Oil Easy to transport and stored
compared to other form (few seasonal) Natural Gas:
Storage cost and transport id high. seasonal fluctua-
tions because of transpot cost amd demand. Elettri-
city: Really difficult to storage, high seasonal, great
variation in price, arbitrage non possible.