ny_quant: (Default)
[personal profile] ny_quant
Мнения жителей бывшей родины на предмет зависимости между стоимостью нефти и курсом рубля к доллару резко разделились. Одни полагают, что "рубль есть дериватив от барреля", другие пренебрежительно смеются. Френд [livejournal.com profile] ammosov опубликовал график взаимного движения этих величин и предложил делать выводы самостоятельно. Хотя его вывод совершенно понятен, в частности из тона записи

Я выразил несогласие с его методами работы с данными и перефигарил по-своему, поменяв координаты местами, обратив курс и посмотрел только на последние 2 года.


Update. В ходе дискуссии зашла речь о скачкообразной смене режима. Вот как выглядят предыдущие 9 месяцев.



Update 2. Пойдя навстречу юзеру [livejournal.com profile] justpeeking4 переделал график в долларовой системе координат.

Date: 2016-01-27 11:07 pm (UTC)
From: [identity profile] ny-quant.livejournal.com
Я думаю, надо строить такие графики, что подсказывают данные. Если в данных находится линейная связь с корреляцией в 99%, то строить надо именно это (а не какие-то непонятные гиперболы). Потому что это, очевидно, неспроста. А потом думать почему так вышло.

Date: 2016-01-28 04:46 am (UTC)
From: [identity profile] ny-quant.livejournal.com
Дык :)

Date: 2016-01-28 10:39 am (UTC)
From: [identity profile] a.s. (from livejournal.com)
I'm not suggesting looking for hyperbolas (USD/OIL vs RUB/USD). I'm suggesting using USD in the denominator of both variables rather than numerator (
as you did). It will preserve (some) linearity but will consider more meaningful variables. Otherwise you are unprincipledly selecting variables of interest which increases meaninglessness of the "results" you find.

Date: 2016-01-28 02:34 pm (UTC)
From: [identity profile] ny-quant.livejournal.com
Some? If y=kx then 1/y = 1/k 1/x

Date: 2016-01-28 06:47 pm (UTC)
From: [identity profile] a.s. (from livejournal.com)
But you are fitting y=kx+a - not y = kx - and the former does not translate into a linear relationship between 1/y and 1/x globally - only locally. Had you fitted y = kx (which would basically correspond to studying RUB/OIL directly), it would be consistent.

Date: 2016-01-28 07:41 pm (UTC)
From: [identity profile] ny-quant.livejournal.com
a is not large in this case but technically you're right

Date: 2016-01-29 12:21 am (UTC)
From: [identity profile] a.s. (from livejournal.com)
a (0.55) is not that small relative to range of y (1.1-3). Based on your graph, RUB/OIL varies from ~2150 (lower left corner) to ~4050 (top right corner) which is almost by a factor of 2. Which should be the case since RUB can't be fully explained by OIL - it'd have staying power even if OIL->0.

Separately, it would be helpful to plot both (slightly smoothed) quantities on the same graph as a function of time since there could be a time-lag in economic effect propagating.
Edited Date: 2016-01-29 01:25 am (UTC)

Date: 2016-01-29 02:00 am (UTC)
From: [identity profile] ny-quant.livejournal.com
Добавил. Заодно пределал первый график, выбросив данные за этот год, там какая-то лажа с данными или опять новая фаза.

Те же яйца, вид сбоку.

Date: 2016-01-29 04:45 am (UTC)
From: [identity profile] a.s. (from livejournal.com)
Upon further inspection, I start to think that USD/OIL is more relevant since Russia sells OIL for USD and then can use that USD to prop by "buying" RUB - so again, we are interested in RUB/USD. Since a specific functional fit is somewhat arbitrary (absent some explicit economic mechanisms), looking at USD/BRT vs RUB/USD makes the most sense and uses the most easily interpretable/recognizable values. What looks like hyperbolas can be also easily approximated by a line if that's your preferred basis.

Here is my contribution which clearly shows a phase shift. I normalized by "stable" 3600 RUB/BRT (mostly based on the second half) and "stable" 30 RUB/USD - mostly based on the first half:


Edited Date: 2016-01-29 05:05 am (UTC)

Profile

ny_quant: (Default)
ny_quant

January 2026

S M T W T F S
    123
45 6 7 8 910
11121314151617
18192021222324
25262728293031

Most Popular Tags

Style Credit

Expand Cut Tags

No cut tags
Page generated Jan. 10th, 2026 09:42 pm
Powered by Dreamwidth Studios