Pair Trading Strategy Journal
- Thread starter jonnysharp
- Start date
I've been pairs trading for several years and as most would proabably agree, most pairs require days or weeks to work. I was wondering, does anybody have any good intra-day money making pairs which can be used daily and are closed out each day? Does the previously mentioned GE/TXT have potential intra-day? Any ideas are greatly appreciated.
intra-day - i look at uso/oih or GLTR/DBC
these pairs look not to bad for intra-day trading
these pairs look not to bad for intra-day trading
----------------------------------------------------------
Quote from OHTodd:
Sell: ALL
Buy: MET
ADF: 5.5
Way to go OHTodd.
Closed this trade today at 8.7%.
Thanks for a good tip.
-----------------------------------------------------------
Dukatu -- sorry, I've been away from the message board for a while. Happy to hear you made the trade, and I hope to post more soon.
Quote from OHTodd:
Sell: ALL
Buy: MET
ADF: 5.5
Way to go OHTodd.
Closed this trade today at 8.7%.
Thanks for a good tip.
-----------------------------------------------------------
Dukatu -- sorry, I've been away from the message board for a while. Happy to hear you made the trade, and I hope to post more soon.
I have seen people post about position sizing in terms of small cap/large cap, dollar sizing, beta-weighting, and standard deviation weighting. I would like to weight-in as a standard deviation weigher.
Everyone understands that trading an equal number of shares between a small cap and a large cap isn't a hedge, but what is?
Beta-weighting is a good strategy to make oneself market neutral, but we can do better with standard deviation weighing.
We have hopefully picked pairs with a real economic basis for being correlated. For example, stocks in the same industry. When I am looking at a pair (usually a futures pair rather than a stock pair,) I want to understand the economic basis of the correlation. Having found a displacement, I want to form an opinion on its validity. For example, suppose company ABC and DEF are equal competitors who have been correlated for decades based on their common dependence on supply and demand, but there have been some recent poor/good management decisions which have driven the prices apart. The current situation (continuance or restoration) then determines if it is a dispersion or revision trade. (I must concur with previous posts that I am only interested in revision trade as they are closing.)
For me, a pair trade is not magical statistical hedge, it is a bet on my understanding of the economics which have driven two related markets apart. I wish to participate in what makes them different, and to hedge against what makes them common.
To understand exactly how this is done, we need to examine correlation. There are two basic things that people do differently when it comes to correlation: basic formulate and input data.
There are two popular methods of computing a correlation coefficient. Most people are using Pearson's (linear) correlation coefficient. Some people use Spearman's (rank) correlation coefficient. Rank correlation does a much better job of identifying related markets which are highly out-of-line. Since we wish to trade small displacement, not bankrupt companies, linear correlation is a better foundation for pair trading. Let us assume linear correlation for the rest of this post.
The other major point of contention is the input data: price itself, daily/monthly difference/return of price, logarithm of price, daily/monthly difference of the logarithm. Since log(x)-log(y) = log(x/y), looking at a logarithmic graph of the ratio of prices is similar to looking at the differences of the logarithm of price. A popular choice on money shows is to use the price itself; because, it makes a simple graphic. Quants look at the daily difference of the logarithm of price which turns out to be a good basis for looking at the daily return of the price. Since daily differences of all type jump around, I use daily differences of the logarithm of price and price to identify correlation and price itself for the trade.
As an aside, others have posted about currency issues. If you want to see your profits in US Dollars, then convert the price of the underlying market to US Dollars first. For example, when trading a JPY denominated future, apply the present-day contract size to obtain the JPY value of a full contract, and then apply the historical currency conversion factor. Some Quants factor back in the currency conversion on purpose as a way to trade an illiquid currency. If it's not highly liquid, I am not interested.
We are trading a pair of highly, consistently correlated markets. If we are making a reversion trade, then we are betting that this will continue. The consistent correlation coefficient tells us that on a volatility(standard deviation) basis, these markets move together. The Person linear correlation coefficient is the inner product of the z-scores:
Here X and Y could be either the price, the daily difference in price, the daily/monthly/quarterly return of price, the logarithm of price, or the daily difference of the logarithm of price for our two markets, respectively.
To balance the trade according to our assumption of consistently correlation, we need to balance according to the standard deviation of the variables correlated. Thus, when trading a small cap against a large cap with half the volatility, by whichever measurement is used, half the position size should be used.
To balance by a constant ratio like 1:1 is not hedging against anything in particular, and should not be considered a hedge.
To balance by price is better in that price tends to be inversely related to volatility.
To balance by beta is better still as it reflects the volatility relative to a common index.
To balance by standard deviation is best; because, we are assuming that the standard-deviation based correlation coefficient will be consistent during our trade. (If not, you don't have a hedge trade.)
Happy Trading
Everyone understands that trading an equal number of shares between a small cap and a large cap isn't a hedge, but what is?
Beta-weighting is a good strategy to make oneself market neutral, but we can do better with standard deviation weighing.
We have hopefully picked pairs with a real economic basis for being correlated. For example, stocks in the same industry. When I am looking at a pair (usually a futures pair rather than a stock pair,) I want to understand the economic basis of the correlation. Having found a displacement, I want to form an opinion on its validity. For example, suppose company ABC and DEF are equal competitors who have been correlated for decades based on their common dependence on supply and demand, but there have been some recent poor/good management decisions which have driven the prices apart. The current situation (continuance or restoration) then determines if it is a dispersion or revision trade. (I must concur with previous posts that I am only interested in revision trade as they are closing.)
For me, a pair trade is not magical statistical hedge, it is a bet on my understanding of the economics which have driven two related markets apart. I wish to participate in what makes them different, and to hedge against what makes them common.
To understand exactly how this is done, we need to examine correlation. There are two basic things that people do differently when it comes to correlation: basic formulate and input data.
There are two popular methods of computing a correlation coefficient. Most people are using Pearson's (linear) correlation coefficient. Some people use Spearman's (rank) correlation coefficient. Rank correlation does a much better job of identifying related markets which are highly out-of-line. Since we wish to trade small displacement, not bankrupt companies, linear correlation is a better foundation for pair trading. Let us assume linear correlation for the rest of this post.
The other major point of contention is the input data: price itself, daily/monthly difference/return of price, logarithm of price, daily/monthly difference of the logarithm. Since log(x)-log(y) = log(x/y), looking at a logarithmic graph of the ratio of prices is similar to looking at the differences of the logarithm of price. A popular choice on money shows is to use the price itself; because, it makes a simple graphic. Quants look at the daily difference of the logarithm of price which turns out to be a good basis for looking at the daily return of the price. Since daily differences of all type jump around, I use daily differences of the logarithm of price and price to identify correlation and price itself for the trade.
As an aside, others have posted about currency issues. If you want to see your profits in US Dollars, then convert the price of the underlying market to US Dollars first. For example, when trading a JPY denominated future, apply the present-day contract size to obtain the JPY value of a full contract, and then apply the historical currency conversion factor. Some Quants factor back in the currency conversion on purpose as a way to trade an illiquid currency. If it's not highly liquid, I am not interested.
We are trading a pair of highly, consistently correlated markets. If we are making a reversion trade, then we are betting that this will continue. The consistent correlation coefficient tells us that on a volatility(standard deviation) basis, these markets move together. The Person linear correlation coefficient is the inner product of the z-scores:
Code:
rho = sum (X - avg(X))/stddev(X) * (Y â avg(Y))/stddev(Y).To balance the trade according to our assumption of consistently correlation, we need to balance according to the standard deviation of the variables correlated. Thus, when trading a small cap against a large cap with half the volatility, by whichever measurement is used, half the position size should be used.
To balance by a constant ratio like 1:1 is not hedging against anything in particular, and should not be considered a hedge.
To balance by price is better in that price tends to be inversely related to volatility.
To balance by beta is better still as it reflects the volatility relative to a common index.
To balance by standard deviation is best; because, we are assuming that the standard-deviation based correlation coefficient will be consistent during our trade. (If not, you don't have a hedge trade.)
Happy Trading
There are few suitable brokers for retail pairs traders.
As an independent pairs trader, you're executing a wholesale strategy but paying retails costs.
The advantage of pairs trading is there is a high degree of predictability. The disadvantage is that the returns are small.
Unless you're ruthlessly focused on transaction costs you will see your capital slowly converted to commissions/margin/haircut - even if you're successful in making profitable trades.
IB is really your only option if you don't want to go the prop route.
With the Volcker rule taking into effect I wld stay away from B/D's
Prop shops only make sense anyway if you are doing =>100shs/Month as a bare minimum. Otherwise, you will be paying $200-395/Month in desk fees as a remote trader. IB's business model is based on offering institutional rates to the retail trader. They have a great short pool, zero or low locate fees, and I have not had any of my short positions bought in.
I wish there were other shops out there offering the same as IB. But right now they are really the only game in town.
Today at open:
Long SAP / Short IBM
what do you think about the trade?
Long SAP / Short IBM
what do you think about the trade?
Looking at the chart:
I see IBM has been trending higher. SAP, starting 10/27 has been trending lower. In at least recent memory, the two stocks track well. Their displacements have resolved in 2-3 months. The current displacement shows no signs of abating. I am not a fan of trading dispersions, but the Z-score could go from the current 2.21 out to 3. If it gets to 4, I would reexamine the fundamentals to see if this displacement represents a game changer.
The current displacement looks similar to the May situation. If the displacement breaks with SAP climbing towards IBM, one could enter a reversion trade with an expectation of 1 ½ months to complete. My back testing found that it is best to take a reversion profit just before they hit parody (say Z-Score 0.1) since often there is some bouncing around zero.
My 2-cents.
I see IBM has been trending higher. SAP, starting 10/27 has been trending lower. In at least recent memory, the two stocks track well. Their displacements have resolved in 2-3 months. The current displacement shows no signs of abating. I am not a fan of trading dispersions, but the Z-score could go from the current 2.21 out to 3. If it gets to 4, I would reexamine the fundamentals to see if this displacement represents a game changer.
The current displacement looks similar to the May situation. If the displacement breaks with SAP climbing towards IBM, one could enter a reversion trade with an expectation of 1 ½ months to complete. My back testing found that it is best to take a reversion profit just before they hit parody (say Z-Score 0.1) since often there is some bouncing around zero.
My 2-cents.