Aspiring Data Science

#books #trading

Ещё немного цитат из Inside the Black Box.

Dark pools are created by brokers or independent firms to allow their customers to trade directly with each other in an anonymous way. They arose in part because of concerns about the market impact associated with large orders. On a dark pool, there is no information provided about the limit order book, which contains all the liquidity being provided by market makers and other participants. Customers are simply posting their orders to the pool and if someone happens to want to do the opposite side of those orders, the orders get filled. As a result of this anonymous process of matching orders, the market is less likely to move as much as it would in a more public venue, where automated market-making practitioners require compensation to take the other side of large orders.

The total cost of transactions for an instrument, holding all else (such as liquidity, trend or volatility) constant, can be visualized as a graph with the size of the order (in terms of dollars, shares, contracts, or the like) on the xaxis and the cost of trading on the y-axis. It is generally accepted by the quant community that the shape of this curve is quadratic, which means that the cost gets higher ever more quickly as the size of the trade gets larger (due to market impact).


While a quadratic cost function is the most commonly agreed upon type, it is not universally accepted. There is good evidence that, at least for U.S. equities, cost per share scales with an exponent of 1.5, or in other words, as the square root of trade size. This empirically seems to be a better fit, while also being supported by the knowledge that many market makers and dealers view their inventory risk as scaling the same way—as the square root of its size.

To mitigate the data-burning form of look-ahead bias, some quant shops take reasonably drastic measures, separating the strategy research function from the strategy selection function and withholding a significant portion of the entire database from the researchers. In this way, the researcher, in theory, cannot even see what data he has and doesn't have, making it much more difficult for him to engage in look-ahead activities. Less draconian, the researcher might simply not be allowed to know or see which data are used for the out-of-sample period, or the portions of data used for in- and out-of-sample testing might be varied randomly or without informing the researcher.


It may turn out that the names the strategy wants to short, and in particular, the most successful short picks, are on hard-to-borrow lists. Hard-to-borrow lists are those stocks that are generally restricted from shorting by the broker, because the broker cannot mechanically locate shares to borrow, which is required in the act of shorting.

Mistakes made during research become baked into a strategy for its lifetime, and then the systematic implementation of this error can become devastating. Moreover, the research effort is not a one-time affair. Rather, the quant must continually conduct a vigorous and prolific research program to produce profits consistently over time.

Models are, by definition, generalized representations of the past behavior of the market. More general models are more robust over time, but they are less likely to be very accurate at any point in time. More highly specified models have the chance to be more accurate, but they are also more likely to break down entirely when market conditions change.

The newest member of the quant-specific risk family is contagion, or common investor, risk. By this, we mean that we experience risk not because of the strategy itself but because other investors hold the same strategies.

The best argument against quant investing is that the markets are quasiefficient, nonlinear, dynamic, and adversarial systems, which makes it extremely hard to forecast asset prices. That said, there's enough empirical evidence in the sustained performance of the best quant funds to soundly refute that this difficulty is impossible to overcome.

www.tgoop.com/AspiringDataScience/2732

69 viewsAnatoly Alekseev, Sep 29 at 03:38

#books #trading

Ещё немного цитат из Inside the Black Box.

Dark pools are created by brokers or independent firms to allow their customers to trade directly with each other in an anonymous way. They arose in part because of concerns about the market impact associated with large orders. On a dark pool, there is no information provided about the limit order book, which contains all the liquidity being provided by market makers and other participants. Customers are simply posting their orders to the pool and if someone happens to want to do the opposite side of those orders, the orders get filled. As a result of this anonymous process of matching orders, the market is less likely to move as much as it would in a more public venue, where automated market-making practitioners require compensation to take the other side of large orders.

The total cost of transactions for an instrument, holding all else (such as liquidity, trend or volatility) constant, can be visualized as a graph with the size of the order (in terms of dollars, shares, contracts, or the like) on the xaxis and the cost of trading on the y-axis. It is generally accepted by the quant community that the shape of this curve is quadratic, which means that the cost gets higher ever more quickly as the size of the trade gets larger (due to market impact).


While a quadratic cost function is the most commonly agreed upon type, it is not universally accepted. There is good evidence that, at least for U.S. equities, cost per share scales with an exponent of 1.5, or in other words, as the square root of trade size. This empirically seems to be a better fit, while also being supported by the knowledge that many market makers and dealers view their inventory risk as scaling the same way—as the square root of its size.

To mitigate the data-burning form of look-ahead bias, some quant shops take reasonably drastic measures, separating the strategy research function from the strategy selection function and withholding a significant portion of the entire database from the researchers. In this way, the researcher, in theory, cannot even see what data he has and doesn't have, making it much more difficult for him to engage in look-ahead activities. Less draconian, the researcher might simply not be allowed to know or see which data are used for the out-of-sample period, or the portions of data used for in- and out-of-sample testing might be varied randomly or without informing the researcher.


It may turn out that the names the strategy wants to short, and in particular, the most successful short picks, are on hard-to-borrow lists. Hard-to-borrow lists are those stocks that are generally restricted from shorting by the broker, because the broker cannot mechanically locate shares to borrow, which is required in the act of shorting.

Mistakes made during research become baked into a strategy for its lifetime, and then the systematic implementation of this error can become devastating. Moreover, the research effort is not a one-time affair. Rather, the quant must continually conduct a vigorous and prolific research program to produce profits consistently over time.

Models are, by definition, generalized representations of the past behavior of the market. More general models are more robust over time, but they are less likely to be very accurate at any point in time. More highly specified models have the chance to be more accurate, but they are also more likely to break down entirely when market conditions change.

The newest member of the quant-specific risk family is contagion, or common investor, risk. By this, we mean that we experience risk not because of the strategy itself but because other investors hold the same strategies.

The best argument against quant investing is that the markets are quasiefficient, nonlinear, dynamic, and adversarial systems, which makes it extremely hard to forecast asset prices. That said, there's enough empirical evidence in the sustained performance of the best quant funds to soundly refute that this difficulty is impossible to overcome.

BY Aspiring Data Science