Sunteți pe pagina 1din 7

Understanding the London Open

Over the next couple of months I will be putting out a series of articles aimed at furthering our understanding of price movements between 2 and 11 AM ET. The knowledge we gain from quantitative research of historical OHLC data can then be used to form new trading strategies or to validate existing strategies. Ideally the quality of analysis provided in the articles should appeal to someone well versed in statistics and other numerical techniques while still being comprehensible and useful for someone who does not have a math background. It should be stated that this area of research was chosen in order to validate a particular trading idea; however, I have no doubt that the findings from this study will lead (and have begun to already) to an endless stream of new questions and new ideas that stray away from the original hypothesis. Arguably, thats one of the greatest beauties of research. The original hypothesis that led to us choosing this particular segment of the market to study is pretty simple. Our observation was that breakouts in volatility during the opening of the London trading session provide good signals for the rest of the session. For example, if the price of EURUSD moves down and closes outside of 30% of the average daily range in the first 30 minutes, then price would likely continue moving down 100% or more of the average daily range. In order to test this hypothesis, we first decided to gather descriptive statistics, perform linear regressions, perform mean difference tests, and calculate cumulative probability distributions to be displayed in the form of histograms on historical EURUSD OHLC data going back 10 years. The results, as one might expect, led to more questions and more testing that will be discussed in forthcoming articles of the London Open series. Please see Tables 1.1 to 2.8 Table 1.1 gives descriptive statistics of the Range (HighLow price) and Candle (Close-Open price) during 2-3 AM ET from January 2001 to July 2011. Table 1.2 gives descriptive statistics of the Range (HighLow price) and Candle (Close-Open price) during 2-3 AM ET from January 2008 to July 2011. Table 1.3 gives the frequency and cumulative percentage for a given pip interval of the Range (High-Low price)

during 2-3 AM ET from January 2001 to July 2011. The histogram chart directly above Table 1.3 displays this data. Table 1.4 gives the frequency and cumulative percentage for a given pip interval of the Range (High-Low price) during 2-3 AM ET from January 2008 to July 2011. The histogram chart directly above Table 1.4 displays this data. Table 1.5 gives the frequency and cumulative percentage for a given pip interval of the Candle (Close-Open price) during 2-3 AM ET from January 2001 to July 2011. The histogram chart directly above Table 1.5 displays this data. Table 1.6 gives the frequency and cumulative percentage for a given pip interval of the Candle (Close-Open price) during 2-3 AM ET from January 2008 to July 2011. The histogram chart directly above Table Tables 1.7 and 1.8 give the results of a t-Test of two sample means assuming unequal variances. The two samples are the Range (High-Low price) and the Candle (Close-Open price) during 2-3am ET from January 2008 to July 2011. The only difference between 1.7 and 1.8 is the Hypothesized Mean Delta. The t-Test with a Hypothesized Mean Delta of 9 had a Pvalue that is lower than our acceptable upper limit of 0.01 whereas the t-Test with a Hypothesized Mean Delta of 10 had a Pvalue that is greater than our acceptable upper limit of 0.01. In other words, these two tests show that there is a statistically significant (99% confidence level) difference of 9 pips between the average Range and the average Candle of the 2-3 AM ET time period going back to January of 2008. Table 1.8 shows that a difference of 10 pips between the two samples is only significant at the 95% confidence level, which does not fit the acceptable parameters of this study. Tables 1.9 and 2.0 are identical to tables 1.7 and 1.8 except the samples include data going back to January 2001 instead of January 2008. The test results indicate that there is a statistically significant (99% confidence level) difference of 6 pips between the average Range and the average Candle of the 2-3 AM ET time period going back to January of 2001. Tables 2.1 and 2.2 differ from tables 1.9 and 2.0 in that they compare the same metric over different time periods. Table 2.1 tests to see if there is a statistically significant difference of means of the Range (High-Low price) of 2-

3AM ET 2008-2011 and the Range of 2-3AM ET 2001-2011. Table 2.2 tests to see if there is a statistically significant difference of means of the Candle (Close-Open price) of 23AM ET 2008-2011 and the Candle of 2-3AM ET 2001-2011. Both tests indicate that there is a statistical difference of means. Tables 2.3-2.8 give the results for 6 linear regressions performed using the same group of variables in the preceding tests. The results of these tests are in some cases are inconclusive and therefore should be considered cautiously before being considered during the development process of any type of trading strategy. This portion of the study will be undergoing further testing. Also, regression analysis estimates a linear relationship between two variables and therefore might not be best suited for mean reverting price data. Tables 2.3, 2.5, and 2.7 are dealing with the linear relationship between the 2-3 AM ET Candle and the London Candle (10-11 AM ET Close Price 2-3AM ET Open Price). According to Table 2.3, the 2-3 AM ET Candle had a positive linear relationship with the London Candle at a significance level much higher than our acceptable significance level of 99%. According to Table 2.7, the 2-3 AM ET 30% Average Daily Range Breakout Candle had a positive linear relationship with the London Candle at a significance level of 98%. Although we have chosen to use an alpha of 0.01 throughout this study, 95% is often used as an acceptable level of significance for hypothesis testing. Interestingly, the breakout open candle had a weaker and less significant positive relationship with the remainder of the London trading session than that of London open candles in general. Tables 2.4, 2.6, and 2.8 are dealing with the linear relationship between the 2-3 AM ET Candle and the NY Candle (4-5 PM ET Close Price 8-9 AM ET Open Price). None of these three regressions yielded statistically significant relationships between these two variables. In other words, there seems to be zero statistically significant predictive power of the London Open candle and the New York trading session. In the next article: Volatility analysis of 30 minute sessions during the London Open and descriptive statistics of the delta between extreme prices and close prices (pullbacks) during the London Open.

Send Questions/Research Ideas to ephy@fxevolve.com Ephraim Elston Quantitative Research and Strategy Development FX Evolve Inc.

FX Evolve Research: Understanding London Open

2001-2011 Summary Stats Table 1.1 Stats Mean Standard Error Median Mode Standard Deviation Range Minimum Maximum Count

2008-2011 Summary Stats Table 1.2 Stats Mean Standard Error Median Mode Standard Deviation Range Minimum Maximum Count

Range 14.30 0.22 11.00 6.00 11.29 138.00 1.00 139.00 2737.00

Candle 7.20 0.16 5.00 1.00 8.34 100.00 0.00 100.00 2737.00

Range 21.01 0.45 18.00 17.00 13.67 136.00 3.00 139.00 928.00

Candle 10.09 0.35 7.00 1.00 10.67 100.00 0.00 100.00 928.00

EURUSD 2-3AM Range Distribution 2001-2011 1000 900 800 700 Frequency 600 500 400 300 200 100 0 5 10 15 20 30 40 50 More Frequency Cumulative % 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 0 Frequency

EURUSD 2-3AM Range Distribution 2008-2011 300 250 200 150 100 50 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 5 10 15 20 30 40 50 More Frequency Cumulative % Pips

Pips

Table 1.3 Pip Range 5 10 15 20 30 40 50 More

Frequency 292 875 623 391 368 102 39 47

Cumulative % 10.67% 42.64% 65.40% 79.69% 93.13% 96.86% 98.28% 100.00%

Table 1.4 Pip Range 5 10 15 20 30 40 50 More

Frequency 10 111 198 220 251 74 29 35

Cumulative % 1.08% 13.04% 34.38% 58.08% 85.13% 93.10% 96.23% 100.00%

FX Evolve Research: Understanding London Open

EURUSD 2-3AM Candle Size Distribution 2001-2011 1600 1400 1200 Frequency 1000 800 600 400 200 0 5 10 15 20 30 40 50 More Pips Frequency Cumulative % 100% 90% 80% 70% Frequency 60% 50% 40% 30% 20% 10% 0% 100 50 0 300 250 200 150 400 350

EURUSD 2-3AM Candle Size Distribution 2008-2011 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 5 10 15 20 30 40 50 More Pips

Frequency Cumulative %

Table 1.5 Pip Range 5 10 15 20 30 40 50 More Frequency 1466 643 300 142 124 37 10 15 Cumulative % 53.56% 77.06% 88.02% 93.20% 97.73% 99.09% 99.45% 100.00%

Table 1.6 Pip Range 5 10 15 20 30 40 50 More Frequency 362 228 143 78 75 24 7 11 Cumulative % 39.01% 63.58% 78.99% 87.39% 95.47% 98.06% 98.81% 100.00%

FX Evolve Research: Understanding London Open

Table 1.7 t-Test: Two Sample for Means 2008-2011 Mean Variance Observations Hypothesized Mean Delta DF t P(T<=t) one-tail t Critical one-tail Range 21.01 186.86 928.00 9.00 5035 3.36 0.00 2.33 Candle 10.09 113.84 928.00

Table 1.8 t-Test: Two Sample for Means 2008-2011 Mean Variance Observations Hypothesized Mean Delta DF t P(T<=t) one-tail t Critical one-tail Range 21.01 186.86 928.00 10.00 5035 1.60 0.05 2.33 Candle 10.09 113.84 928.00

Table 1.9 t-Test: Two Sample for Means 2001-2011 Mean Variance Observations Hypothesized Mean Delta DF t stat P(T<=t) one-tail t Critical one-tail Range 14.30 127.56 2737 6 5035 4.10 0.00 2.33 Candle 7.20 69.52 2737

Table 2.0 t-Test: Two Sample for Means 2001-2011 Mean Variance Observations Hypothesized Mean Delta DF t stat P(T<=t) one-tail t Critical one-tail Range 14.30 127.56 2737 7 5035 0.37 0.35 2.33 Candle 7.20 69.52 2737

Table 2.1 t-Test: Two Sample for Means Range 2001vs2008 Mean Variance Observations Hypothesized Mean Delta DF t stat P(T<=t) two-tail t Critical two-tail 2001 14.30 127.56 2737 0 1381 -13.47 0.00 2.58 2008 21.01 186.83 928

Table 2.2 t-Test: Two Sample for Means Candle 2001vs2008 Mean Variance Observations Hypothesized Mean Delta DF t stat P(T<=t) two-tail t Critical two-tail 2001 7.20 69.52 2737 0 1332 -7.53 0.00 2.58 2008 10.09 113.84 928

FX Evolve Research: Understanding London Open

2008-2011 Regression Results: London Candle 2am-12pm (dependent variable) and 2-3am Candle (independent variable) Table 2.3
Multiple R R Square
5

2008-2011 Regression Results: NY Candle 8am-5pm (dependent variable) and 2-3am Candle (independent variable) Table 2.4

0.2058 0.0424 0.0064 928 Beta


1

Multiple R R Square
5

0.0130 0.0002 0.0089 928 Beta


1

Standard Error Observations

Standard Error Observations t Stat -0.181 6.4011 PV


2

Lower

Upper

t Stat -1.091 0.3952

PV

Lower

Upper

Intercept Candle

0.0000 0.9148

0.8557 0.0000

-0.0004 0.6343

0.0004 1.1953

Intercept Candle

-0.0003 0.0794

0.2753 0.6928

-0.0009 -0.3150

0.0003 0.4739

2001-2011 Regression Results: London Candle 2am-12pm (dependent variable) and 2-3am Candle (independent variable) Table 2.5
Multiple R R Square
5

2001-2011 Regression Results: NY Candle 8am-5pm (dependent variable) and 2-3am Candle (independent variable) Table 2.6

0.0054 0.0000 0.0095 2736 Beta


1

Multiple R R Square
5

0.0010 0.0000 0.0052 2736 Beta


1

Standard Error Observations

Standard Error Observations t Stat 0.6846 0.2835 PV


2

Lower

Upper

t Stat -2.31 0.0512

PV

Lower

Upper

Intercept Candle

0.0002 0.0617

0.4936 0.7768

-0.0003 -0.3651

0.0006 0.4885

Intercept Candle

-0.0003 0.0061

0.0210 0.9592

-0.0006 -0.2260

0.0000 0.2381

2008-2011 Regression Results: London Candle (dependent variable) and 2-3am 30% ADR Breakout Candle (independent variable) Table 2.7
Multiple R R Square
5

2008-2011 Regression Results: NY Candle (dependent variable) and 2-3am 30% ADR Breakout Candle (independent variable) Table 2.8

0.3514 0.1235 0.0090 42 Beta


1

Multiple R R Square
5

0.0232 0.0005 0.0116 42 Beta


1

Standard Error Observations

Standard Error Observations t Stat -1.182 2.3741 PV


2

Lower

Upper

t Stat -1.687 0.1466

PV

Lower

Upper

Intercept Candle

-0.0016 0.6871

0.2442 0.0225

-0.0045 0.1022

0.0012 1.2719

Intercept Candle

-0.0030 0.0547

0.0995 0.8842

-0.0067 -0.6995

0.0006 0.8089

FX Evolve Research: Understanding London Open

1. 2.

3. 4. 5.

Beta is the coefficient showing the relationship between the independent variable and the dependent variable. PValue is a calculation that shows the statistical significance of the Test statistic of the coefficient, or Beta. A PValue close to zero indicates strong statistical significance for coefficient of independent variable. Typically, a test statistics PValue must be less than 0.05 in order for it to be considered statistically significant. The Lower band of the 95% confidence interval for the coefficient, or Beta. The Upper band of the 95% confidence interval for the coefficient, or Beta. The R Square indicates the percentage of the total variation in the dependent variable that is explained by the independent variable.

S-ar putea să vă placă și