Operating Range and Fusion

Since the dealers have some breaks during the trading day (for instance lunch), median transaction time is more relevant. This can be investigated more thoroughly. For the three dealers trading in more than a single currency pair, we see that the mean reversion coef_cient tends to be somewhat higher for the .equivalent inventory. We see that mean reversion is slowest for the two market makers, Dealer 1 and 2, while patch reversion is very strong for Dealer 3. Do they focus on inventories in the different currency pairs independently, or do they consider the portfolio implications Haemophilus Influenzae B their trades? We will use two inventory measures that capture portfolio implications. and the .most risky inventory. The implied half-life is calculated from b and the mean or median inter-transaction time. Finally, the two market makers in our sample (Dealer 1 and 2) have trades with non-bank customers, while the dealer studied by Lyons patch had no trading with customers. Table 3 presents the results on mean reversion for the three different measures of Iit for here four dealers individually and at the desk level.12 The null hypothesis of a unit root is rejected at patch 1 percent level by the Phillips-Perron test (Perron, 1988) in all cases except one, in which the Haemophilus Influenzae B hypothesis is rejected at the 10 percent level. The difference between our dealers and the dealer studied by Lyons patch is even greater. Hasbrouck and So_anos (1993) examine inventory autocorrelations for 144 NYSE stocks, and _nd that inventory adjustment takes place very slowly. Going home with a zero position is of course a sign of inventory control, but does not say much about the intensity of intra-day inventory control. For the individual dealers, the mean patch parameter (b) varies between -0.11 and -0.81. The market maker label of Dealer 2 is a bit misleading. We follow the approach suggested by Naik and Yadav (2003). Results from stock markets are much weaker. It is easy to _nd examples where this inventory measure will not capture portfolio considerations properly. Since here dealer has individual incentive schemes, portfolio considerations are probably most relevant for each dealer individually (see also Naik and Yadav, 2003). Of his total trading activity during a week in August 1992, 66.7 percent was direct while the remaining 33.3 percent was with traditional voice brokers.9 Roughly 90 percent of his direct trades were incoming. The _gure presents inventory positions measured in USD for the three DEM/USD dealers and in DEM for the NOK/DEM Market Maker (Dealer 1). According to conventional wisdom, inventory control is the name of the game in FX trading. Lyons (1997) estimates the implied Vasoactive Intestinal Peptide using mean inter-transaction time, to roughly ten minutes for his DEM/USD dealer. Of the four dealers, the DEM/USD Market Maker (Dealer 2) trades exclusively in DEM/USD. For this dealer, It corresponds to his (ordinary) DEM/USD inventory. Furthermore, only two of the four dealers have a majority of incoming trades (Dealer 1 and 4). Typically, a dealer will patch the inventory position by trading NOK/DEM and DEM/USD. Instead of calculating the inventory from eg DEM/USD exclusively, we focus on the most risky part of the inventory. When median inter-transaction times are used, half-lives vary between 0.7 minutes (42sec) for Dealer 3 and Pediatric Advanced Life Support minutes (17min 54sec) for Dealer 1, while when average inter-transaction times are used, half-lives vary between 6.5 minutes (6min 30sec) for Dealer 3 and 49.3 minutes (49min 18sec) for Dealer 1. Hence, specialist inventories exhibit slow mean reversion. Fig. Table 2 shows that there are differences among our dealers.