Financial Economics Week 3 Introduction Filippo Massari University of Bologna 23 Feb 2024 Filippo Massari (University of Bologna) Week 3 22/4 1 / 38 The efficient frontier with many assets Suppose there are N risky assets (where N can be arbitrarily large) A portfolio of assets is a vector (we shall define vectors as column vector and indicate row vectors by the transpose symbol X′ ) X′ = [X1 , X2 , ..., XN ] where the X s are the fraction of wealth invested in each asset If short sales are allowed, some of the X s can be negative (of course, not all of them, since the sum must be 1) Each asset i has an expected return R̄i The vector of expected returns is R′ = [R̄1 , R̄2 , ..., R̄N ] Filippo Massari (University of Bologna) Week 3 22/4 2 / 38 Many assets The matrix of variances and covariances (for brevity, covariance matrix) is 2 σ1 σ12 σ13 ... σ1N σ21 σ22 σ23 ... σ2N 2 Σ= σ31 σ32 σ3 ... σ3N ... ... ... ... ... σN1 σN2 σN3 ... σN2 With N assets, the variance-covariance matrix has N 2 entries. N (N + 1) However, σ12 = σ21 etc. so the independent entries are . 2 The inputs for the portfolio selection process (in addition to the N (N + 1) investor’s preferences over risk) are variances and 2 covariances plus the N expected returns With 500 assets, the inputs are about 125,000! Filippo Massari (University of Bologna) Week 3 22/4 3 / 38 Many assets We assume that all assets are risky In matrix notation, the expected return of a portfolio is R̄P = X′ R The variance of the portfolio rate of return is σP2 = X′ ΣX One can easily verify that in the two asset case one re-obtains the formulas of Lecture 2 Note : the covariance matrix is symmetric and positive defined, so X′ ΣX is a quadratic form In our analysis we shall assume that Σ has full rank (is not singular, has a determinant different from 0) The economic interpretation is that there are no risk free assets and no asset is perfectly correlated with a combination of the others Filippo Massari (University of Bologna) Week 3 22/4 4 / 38 Many assets: a simple case In general the variance-covariance matrix can be very complicated A simple special case is that all variances are equal and also that all covariances are equal (but different from the variances) In this case the covariance matrix becomes 2 σ ρσ2 ... ρσ2 ρσ2 σ2 ... ρσ2 Σ = ... ... ... ... ρσ2 ρσ2 ... ρσ2 1 ρ ... ρ ρ 1 ... ρ = σ2 ... ... ... ... ρ ρ ... ρ Filippo Massari (University of Bologna) Week 3 22/4 5 / 38 Many assets: a simple case Assume, in addition, that 1 1 1 X = , , ..., N N N ′ That is, the same fraction of wealth is invested in each of the N assets In this case the variance of the portfolio becomes σP2 2 1 2 2 1 σ + N (N − 1) ρσ2 N N N −1 σ2 ρσ2 + N N = N = Filippo Massari (University of Bologna) Week 3 22/4 6 / 38 Many assets: a simple case σP2 = σ2 + N N −1 N ρσ2 The first term is the contribution to the portfolio risk of the riskiness of the individual assets (idiosyncratic risk – also called diversifiable risk, firm-specific risk etc.), the second term is the contribution to the portfolio variance of the various assets being correlated (nondiversifiable risk – also called market risk, systematic risk etc.) As N increases, the first term vanishes and the second term converges to ρσ2 That is, idiosyncratic risk can be eliminated by diversification, but nondiversifiable risk cannot Filippo Massari (University of Bologna) Week 3 22/4 7 / 38 Many assets: a simple case The figure shows how the risk of the portfolio decreases as the number of securities included increases using average variance and covariance of the relevant markets. Filippo Massari (University of Bologna) Week 3 22/4 8 / 38 The efficient frontier with many assets One way to calculate the efficient frontier is to solve the problem = X′ ΣX s.t. X′ R = R̄P (a given constant) X′ 1 = 1 min σP2 where 1 is a vector of ones, for different values of Rp (Merton, JFQA, 1972, Roll, JFE 1977) An alternative but equivalent method is to solve the problem 1 1 max t R̄P − σP2 = tX′ R− X′ ΣX 2 2 ′ s.t. X 1 = 1 for different values of t (Szego 1974) The parameter t is called tolerance for reasons that will become clear when we study the investor’s preferences For each value of R̄P (or t) we find one point of the efficient frontier; by varying R̄P (or t) we can trace out the entire frontier Filippo Massari (University of Bologna) Week 3 22/4 9 / 38 The efficient frontier with many assets We start from the second method Since we have a constrained maximisation problem, we set up the Lagrangian 1 L = tX′ R− X′ ΣX+λ(1 − X′ 1) 2 where λ is a Lagrange multiplier First order conditions for a maximum are (the derivative of a quadratic form is similar to the derivative of ax 2 ) tR−ΣX−λ1 = 0 X′ 1 = 1 The solution is X =Σ−1 (tR−λ1) Filippo Massari (University of Bologna) Week 3 22/4 10 / 38 The efficient frontier with many assets There are two parameters in the solution However, one is pinned down by the adding up condition X′ 1 = 1′ X = 1 We can get rid of the redundant parameter by noting that λ −1 X =tΣ R− 1 t Denoting λ = c we have t Σ−1 (R−c1) X = ′ −1 1 Σ (R−c1) By varying c we can now trace out the entire frontier Filippo Massari (University of Bologna) Week 3 22/4 11 / 38 The efficient frontier with many assets Using the constraint 1′ X = 1 to eliminate λ we get B −1 1 −1 −1 X = Σ 1 + t Σ R− Σ 1 C C where C = 1′ Σ −1 1 and B = R′ Σ −1 1 are scalars Clearly, X is linear in t Note that we are allowing for short sales and hence some of the X s can be negative Define also A = R′ Σ−1 R and D = AC − B 2 Filippo Massari (University of Bologna) Week 3 22/4 12 / 38 The efficient frontier with many assets Given the solution we have found, we can calculate RP = R′ X B2 B + t (A − ) = C C 1 = (B + Dt ) C and σP2 = X′ ΣX = Filippo Massari (University of Bologna) Week 3 Dt 2 + 1 C 22/4 13 / 38 The efficient frontier with many assets From these conditions we easily obtain σP2 = A − 2BRP + CRP2 D This is the equation of a parabola In terms of the standard deviation s A − 2BRP + CRP2 σP = D which is the equation of a hyperbola Filippo Massari (University of Bologna) Week 3 22/4 14 / 38 The efficient frontier with many assets Consider now the other procedure, which is equivalent to = X′ ΣX s.t. X′ R = R̄P X′ 1 = 1 max −σP2 The Lagrangian is 1 L = − X′ ΣX+µ(R̄P − X′ R)+λ(1 − X′ 1) 2 where now we have two Lagrange multipliers, λ and µ But this is equivalent to the Lagrangian we have used before setting µ = t (they differ only by a constant term µR̄P ) In fact µ will depend on R̄P and so must vary arbitrarily, just as t does This shows that the two methods are equivalent Filippo Massari (University of Bologna) Week 3 22/4 15 / 38 The two mutual funds theorem We can now prove one of the most elegant results of portfolio theory: the two mutual funds theorem (also known as the two funds theorem, the mutual funds theorem, or the separation theorem) The theorem says that the efficient frontier can be generated by combining any two portfolios on the frontier, just as we combined single assets in Lecture 1 This implies that all an investor has to do is to calculate two efficient portfolios; any other efficient portfolio then is a combination of these two (possibly with negative weights for one portfolio) Thus the problem of finding the optimal portfolio becomes the problem of finding the optimal weights We shall see later that when there is a risk free asset the two mutual funds theorem has even striker implications Filippo Massari (University of Bologna) Week 3 22/4 16 / 38 The two mutual funds theorem Proof: The trick is to note that X is linear in t B −1 1 −1 −1 Σ 1 + t Σ R− Σ 1 X = C C = M + tN Take two different values of t, t1 ̸= t2 , (say t1 > t2 ) and consider the corresponding efficient portfolios X1 = M + t1 N X2 = M + t2 N Solve this system for M and N N = M = Filippo Massari (University of Bologna) X1 − X2 t1 − t2 t 1 X2 − t 2 X1 t1 − t2 Week 3 22/4 17 / 38 The two mutual funds theorem Given the solution for M and N N = M = X1 − X2 t1 − t2 t 1 X2 − t 2 X1 t1 − t2 we can express any efficient portfolio X as a combination of X1 and X2 (t1 − t )X2 −(t2 − t )X1 X = M + tN = t1 − t2 Filippo Massari (University of Bologna) Week 3 22/4 18 / 38 Risk-free asset Now suppose that there is a risk free asset (if there were many they should all have the same return, otherwise – if they are really risk-free – there would be arbitrage opportunities) In the past, the prototypical example of risk-free assets would have been a government bond Today it is apparent that even government bonds can be risky for many countries Nevertheless, one can still think of government bonds of such countries as US or Germany as risk free With a risk free asset, the covariance matrix is singular (there will be one row and one column of 0s) so we cannot apply the above analysis However, it turns out that the analysis is in fact even simpler Filippo Massari (University of Bologna) Week 3 22/4 19 / 38 Two risky assets and one risk-free asset Recall that we can combine not only individual assets but also portfolios of assets In particular, we can combine any portfolio on the risk-return frontier with the risk free asset As we have shown, the expected return/standard deviation for any portfolio which is the combination of a risky asset or portfolio and the risk free asset must lie on the straight line connecting the risky portfolio to the risk free asset Going short on the risk free asset is equivalent to borrowing at the risk free interest rate (something that individuals and even many governments cannot in fact do). Filippo Massari (University of Bologna) Week 3 22/4 20 / 38 Two risky assets and one risk-free asset Clearly, there will be one efficient risky portfolio, the tangency portfolio. The efficient frontier will be a mixture of the risk-free asset and the tangency portfolio Filippo Massari (University of Bologna) Week 3 22/4 21 / 38 The separation theorem From the figure above we can draw a simple conclusion Because the efficient frontier is a mixture of the risk free asset and the optimal risky portfolio, risk preferences do not determine which risky portfolio is chosen, but only the proportion of the risk-free asset and the efficient risky portfolio This conclusion, first noticed by Nobel laureate James Tobin (REStat 1958), is called the separation theorem (or sometimes, more modestly, the separation property) It may be seen as a special case of the two mutual funds theorem and implies that the selection of the optimal portfolio can be divided into two steps: finding the optimal risky portfolio finding the share of wealth to be invested in the risky portfolio The investor’s preferences come into play only at the second stage Filippo Massari (University of Bologna) Week 3 22/4 22 / 38 The separation theorem The optimal risky portfolio will of course depend on the investor’s assessment of the expected return, the variance of returns and the covariance of returns of the risky assets However, it does not depend on the investor’s preferences over risk and return This implies that if you have many different clients, with different attitudes towards risk, you will recommend different combinations of the risk free asset and the optimal risky portfolio However, you will recommend the same risky portfolio to all of your clients, irrespective of their risk attitude Filippo Massari (University of Bologna) Week 3 22/4 23 / 38 The Capital Allocation Line (CAL) The capital allocation line (CAL) is the line that joins the risk free asset and the optimal risky asset It must be tangent to the efficient frontier Filippo Massari (University of Bologna) Week 3 22/4 24 / 38 The Sharpe ratio The Sharpe ratio is the slope of the capital allocation line It is given by R̄P − RF Sharpe ratio = σP Risk premium = Volatility where R̄P is the expected rate of return of the optimal risky portfolio and σP is its standard deviation The numerator of the Sharpe ratio can be interpreted as a risk premium (the excess return which is obtained by holding a risky portfolio); the denominator as a measure of volatility of the portfolio which yields excess returns This reward to volatility measure was first proposed by William Sharpe, who shared with Markowitz the Nobel prize in 1990 Filippo Massari (University of Bologna) Week 3 22/4 25 / 38 The optimal risky portfolio The optimal risky portfolio is the one with the highest Sharpe ratio To find it, one must maximise θ = R̄P − RF σP among all feasible portfolios of risky assets Clearly, the solution will necessarily lie on the efficient frontier Filippo Massari (University of Bologna) Week 3 22/4 26 / 38 The optimal risky portfolio Recalling the expressions for R̄P and σP , our problem is to maximise θ = = R̄P − RF σP X1 (R̄1 − RF ) + X2 (R̄2 − RF ) q X12 σ12 + X22 σ22 + 2X1 X2 σ12 We need not impose the constraint X1 + X2 = 1 as the objective function is homogeneous of degree 0 in the X s Filippo Massari (University of Bologna) Week 3 22/4 27 / 38 The optimal risky portfolio The first order conditions are dθ dX1 dθ dX2 = 0, = 0 which are equivalent to 1 2X1 σ12 + 2X2 σ12 (R̄1 − RF ) σP − (R̄P − RF ) 2 σP 1 2X2 σ22 + 2X1 σ12 (R̄2 − RF ) σP − (R̄P − RF ) 2 σP Filippo Massari (University of Bologna) Week 3 = 0, = 0 22/4 28 / 38 The optimal risky portfolio Rearranging R̄P − RF 2 X σ + X σ = 0, 1 2 12 1 σP2 R̄P − RF X2 σ22 + X1 σ12 = 0 (R̄1 − RF ) − 2 σP (R̄1 − RF ) − Define new variables Z1 = R̄P − RF X1 σP2 Z2 = R̄P − RF X2 σP2 This trick works because all we actually need to find are the shares X1 X1 +X2 , which are the same for the X s and the Z s Filippo Massari (University of Bologna) Week 3 22/4 29 / 38 The optimal risky portfolio With this substitution, the system becomes Z1 σ12 + Z2 σ12 = (R̄1 − RF ) Z1 σ12 + Z2 σ22 = (R̄2 − RF ) The solution is Z1 = Z2 = Filippo Massari (University of Bologna) (R̄1 − RF ) σ22 − (R̄2 − RF ) σ12 2 σ12 σ22 − σ12 (R̄2 − RF ) σ12 − (R̄1 − RF ) σ12 2 σ12 σ22 − σ12 Week 3 22/4 30 / 38 The optimal risky portfolio With the Z s, we can then obtain the X s as follows: X1 = (R̄1 − RF ) σ22 − (R̄2 − RF ) σ12 (R̄1 − RF ) (σ22 − σ12 ) + (R̄2 − RF ) (σ12 − σ12 ) and X2 = 1 − X1 Filippo Massari (University of Bologna) Week 3 22/4 31 / 38 Optimal risky portfolio with many assets However large the number of assets, the efficient frontier in the (σP , RP ) space is a hyperbola Above the minimum variance portfolio, the frontier is necessarily concave (the geometric arguments used in the two asset case continue to hold with no change in the N asset case) Thus, the optimal risky portfolio is again the one that maximises the Sharpe ratio θ = = = Filippo Massari (University of Bologna) R̄P − RF σP XR − RF √ X′ ΣX ′ X (R−RF 1) √ X′ ΣX Week 3 22/4 32 / 38 Optimal risky portfolio with many assets The first order condition is (again we need not worry about the constraint X′ 1 = 1) (R−RF 1) √ X′ ΣX − or (R−RF 1) − Filippo Massari (University of Bologna) 1 X′ (R−RF 1) 2ΣX √ =0 2 X′ ΣX X′ (R−RF 1) ΣX = 0 X′ ΣX Week 3 22/4 33 / 38 Optimal risky portfolio with many assets Denoting X ′ ( R − RF 1 ) X′ ΣX (a scalar) and λX = Z,the system reduces to λ= ΣZ = (R−RF 1) The solution is Z = Σ − 1 ( R − RF 1 ) The X s can now be easily recovered as X = = Filippo Massari (University of Bologna) 1 Z 1′ Z Σ−1 (R−RF 1) 1 ′ Σ − 1 ( R − RF 1 ) Week 3 22/4 34 / 38 Optimal risky portfolio: no short sales The solution we have found typically involves Xi ̸= 0; however, Xi can be either positive or negative A negative value of Xi means that you should go short on asset i If short sales are prohibited, one must add the constraint X ≥ 0 in the optimisation problem considered above A closed form solution is no longer available, although all the main qualitative results we have obtained (namely, the shape of the frontier and the separation property) continue to hold However, the relationship between the efficient frontier with and without short sales is not as simple as in the two assets case With no short sales, not only does the frontier not extend indefinitely: it also lies everywhere below the frontier with short sales Filippo Massari (University of Bologna) Week 3 22/4 35 / 38 Borrowing constraints If the investor cannot borrow at the risk free rate, the separation theorem no longer holds The efficient frontier becomes itbpF 2.2477in2.4785in0inFigure Filippo Massari (University of Bologna) Week 3 22/4 36 / 38 Borrowing constraints If the investor can only borrow at an interest rate higher the risk free rate, the efficient frontier will be Filippo Massari (University of Bologna) Week 3 22/4 37 / 38 Borrowing constraints With borrowing constraints, we have a three mutual funds theorem That is, the efficient frontier can be generated by combing three different portfolios: the risk free asset (which has a different rate of return depending on whether you go short or long), portfolio G, and portfolio H Filippo Massari (University of Bologna) Week 3 22/4 38 / 38