Discussion: liquidity Creation as Volatility Risk by Itamar Drechsler, Alan Moreira, and Alexi Savov Yunzhi Hu University of North Carolina FMA, Nov 9, 2018 1/7
Summary Motivation: liquidity and volatility 1. Transaction (participation/inventory) cost: Nagel (2012) 2. Asymmetric information: this paper Main idea: A Kyle (1985) static model with stochastic volatility ◮ All orders flow in at date 1; asset pays off at date 1 p i , 1 = constant + σ i , 1 v i ���� ���� private information sto. vol ◮ Informed traders’ information is more (less) valuable during periods of high (low) realized volatility. ◮ By providing liquidity, market makers have negative exposure to volatility risk. ◮ Asset-level volatility is highly correlated with aggregate volatility. Hence volatility risks cannot be diversified away. ◮ Liquidity creation demands a premium. 2/7
Summary ◮ Empirics • Liquidity providers’ positions ≈ short-term reversal portfolios • Short-term reversal strategies have negative exposure to volatility shocks ( ≈ ∆ VIX ) • The five-day large stock reversal return drops by 64 bps if VIX rises by an average of one-point per day over the holding period. • The impact persists for (at least) five days. • Volatility risk exposure is priced and accountable for the average returns of the reversal strategies. 3/7
Comment 1: robust evidence 1. This paper implies that liquidity creation suffers losses during periods of high realized volatility. Do we see this in the data? � � R p Corr t , t +5 , RV t , t +5 ≶ 0 2. Days before earnings announcement versus other periods? 3. NYSE versus NASDAQ? 4. Dealer driven versus algorithm driven? • Split the sample by 2005 • Menkveld (2016): HFTs avoid carrying a position overnight. Use (close price - open price) to calculate returns? 4/7
Comment 2: measurement ◮ Which returns? • Bid-ask bounce? Use quote midpoint changes? • Hedged returns or raw returns? Market makers hedge common factor exposures with S&P 500 futures contracts. ◮ ∆ VIX or ∆ VIX t VIX t − 1 ? • The volatility of volatility ◮ Portfolio formation: why weighted by dollar volume • In Lehmann (1990) and Nagel (2012), the weights predicted by the model are derived from previous period’s returns. • Campbell, Grossman, and Wang (1993) suggests illiquidity should be measured by return autocovariance conditional on trading volume. Llorente, Michaely, Saar, and Wang (2002) show that conditional γ correlates negatively with measures of asymmetric information. 5/7
Comment 3: effect persistence ◮ Is the persistence driven by persistence in ∆ VIX ? ∆ VIX t = − 0 . 1114 (0 . 0157) ∆ VIX t − 1 + ε t ◮ When do the effects finally disappear? • When is private information revealed in price? With multiple insiders, information revelation slows down. (Foster and Viswanathan (1996), Back, Cao, and Willard (2000)) • Half-life of dealer inventory: 2.5 days in Hansch, Naik, and Viswanathian (1998); 0.92 days in Hendershott and Menkveld (2014) 6/7
Comment 4: short-term reversal strategies ◮ Can returns to reversal strategies capture more than returns to liquidity creation? • Sentiment: fads, overreaction, cognitive errors ◮ Are negative β s driven by buying losers or selling winners? • Da, Liu, and Schaumburg (2014): buying losers load on illiquid measures (Amihud and realized volatility of S&P 500 index); selling winners load on lagged investor sentiment (IPOs and equity issuance) • Stambaugh and Yuan (2017): investor sentiment predicts the short legs of mispricing factors 7/7
Recommend
More recommend
