This page describes the research project undertaken by forexhit.com to uncover the** correlation periods** that are best suited for monitoring the short-term and the long-term daily correlations between the currency pairs.The aim of this project was to find out if there are correlation periods (short-term and long-term) that are consistently better than the others in terms of their correspondence with the day-to-day price fluctuations.

The **correspondence** of a correlation period with the daily price fluctuations was measured by taking the average of the daily **differences between the correlation coefficient and the coefficient of the proportionality. **

The correlation coefficient was calculated for N days back starting from the **preceding day **(N being the length of the correlation period). For example, when N was 50, the correlation coefficient was calculated using the closing prices of the last 50 trading days starting from yesterday's close.

The coefficient of proportionality was calculated by **comparing the distances** between the levels of Open, High, Low and Close that the two pairs showed during each trading day. The purpose of this coefficient was to determine the degree of **actual** correspondence between the movements of two pairs from the **relative** pip distances between the four price levels that are recorded in their daily candlesticks. The coefficient of proportionality can also be thought of as a **proxy for the intraday correlation between the pairs** - or as a method to estimate the intraday correlation between the two pairs by using only the levels recorded in their daily candlestick and **not** bringing up the hourly or the 30-minute charts for the same day. The coefficient of proportionality was, therefore, calculated in such a way so as to be easily comparable to the coefficient of correlation. Taking the daily difference between these two coefficients allowed, in essence, to measure **how well the daily correlation **calculated over N immediately preceding days corresponded with the** intraday correlation of the next day.**

Both coefficients and their difference were calculated for each trading day for the last 4 years. Then, the average was taken of all the daily differences between these two coefficients. The smaller the average difference between the coefficients is, the better the correlation period predicts the relative price movement of the two pairs in the **immediate future** (the next day) and, therefore, the more useful it is for **practical** application.

The above calculation was** looped through for 197 different correlation periods **(starting from 3 days to 200 days) for fifteen different currency pair combinations. For each currency pair combination a scatter chart was created which plotted the level of the correspondence of the correlation period to the daily price movement for each of the 197 periods that were tested. The following are the currency pair combinations that were used in this analysis:

1 | USDCAD&EURUSD |

2 | USDCAD&USDJPY |

3 | USDCAD&GBPUSD |

4 | USDCAD&USDCHF |

5 | AUDUSD&USDCAD |

6 | AUDUSD&EURUSD |

7 | AUDUSD&USDJPY |

8 | AUDUSD&GBPUSD |

9 | AUDUSD&USDCHF |

10 | GBPUSD&USDJPY |

11 | EURUSD&USDJPY |

12 | EURUSD&GBPUSD |

13 | USDJPY&USDCHF |

14 | EURUSD&USDCHF |

15 | GBPUSD&USDCHF |

Note: You can download all the charts that were created in this project in one PDF file - correlation periods.pdf (requiresAdobe Acrobat Reader).

Once the all the charts have been plotted it became apparent that **most of the correspondence curves reached their minimums around two periods - of 40 and 120 days**. This tendency became more clear when the 15 curves were overlaid on a single chart. Below are the two overlay charts - one shows only the correspondence curves while the other adds correlation period labels to the points on the correspondence curves. Because all of the pairs include US Dollar, the names of pair combinations in these charts were shortened by excluding the "USD" from the pairs' names.

On both overlay charts the 40 and 120 correlation periods were marked with black vertical lines. These lines make it easy to see that most correspondence curves** start to rise** after they pass these correlation periods (i.e. the correspondence weakens). To confirm that this was indeed the general tendency an average curve was calculated from all the 15 curves (black squares on the above charts) and its points sorted in an ascending order - starting from the best correspondence (minimum average difference between the daily coefficients of proportionality and correlation). Following are the 20 correlation periods which correspond to the first 20 points in the list (which has a total of 197 points in it).

1 | 38 |

2 | 39 |

3 | 37 |

4 | 40 |

5 | 36 |

6 | 41 |

7 | 34 |

8 | 35 |

9 | 33 |

10 | 42 |

11 | 32 |

12 | 43 |

13 | 31 |

14 | 120 |

15 | 44 |

16 | 121 |

17 | 119 |

18 | 122 |

19 | 118 |

20 | 117 |

As can be readily seen from the above table, correlation periods around the 40-day and 120-day marks are indeed generally better than others in tying the daily correlation to the actual daily price activity. This is reason why the 40-day period is used to monitor the shorter-term currency correlations and the 120-day is used to monitor the longer-term currency correlations on this site. For details on the recent currency correlations please visit the currency correlations or the trailing correlations webpage. (Please note: The size of this page is 1,3 Mbs and it requires that you have Flash installed and Javascript enabled in your browser)