In sivadevops1/playbook: Playbook is the abstract on top of rulesNtrades, dataPrepare and evalStrats.

Gold Data

library(Quandl)
library(dplyr)

## --- Get Diwali days -----------------------------------------------------------------
d <- read.csv("E:/R/Projects/playbook/diwaliDates",header = FALSE,stringsAsFactors = FALSE)
d <- strptime(d[,1], format = "%d %B %Y")
d <- as.Date(d)

## --- Get goldbees price----------------------------------------------------------------
gb <- Quandl("NSE/GOLDBEES",collapse="daily",start_date="2005-01-01",type="raw")
gb <- xts(gb[c("Open","High","Low","Close")],order.by = gb$Date)

Diwali inflates the gold prices - Really ?

Here we studied the gold prices how they behaved 'N' days prior to the big Indian festive event - Diwali, the usual period where everyone buys the gold. So, if the demand for gold increases during festive season, so should be the prices. A comparision of prices from past 'N' days till diwali is studied. Unfortunately, the prices around 75% of the time are less than 60 percentile when it nears diwali. So, is it not Diwali that is causing the price fluctuations. Then who ?

Another study done comparing the prices with international prices & correlation. The Indian gold price behaviour is almost inclined (90% of the time) to the international behaviour(where Diwali demand doesn't exist) rather than local demand - supply.

Its demand that definitely reason behind price action, but the local demand is a miniscule when compared to the factors like interest rates, volatility in the world markets, US Dollar strength, hedging, inflation & interest rate cycles.

For Study - 1. Two Gold ETF prices (one from India and one from US) are taken from 2008.
a. GLD - Managed by SPDR Trust in US
b. GOLDBEES - Managed by reliance in India
2. The prices prevN (60 days,30 days) of each year are considered as subset
3. To see if prices are increasing towards Diwali, close of price (percentile/quartile) with respect to the High-low price is measured. More the percentile, the prices are inclined towards Highs.
4. To see if prices are correlated with international prices (GLD), statistical correlation is considered

Diwali Days

cat(format(d,"%Y %B %d"),sep = "#  \n")

prevN <- 60
yearlyData <- lapply(d,
                     FUN = function(x){
                       tmp <- (x-prevN):(x-2)
                       gbXts[as.Date(tmp)]
                       }
                     )

Goldbees price during Diwali for '60' previous days

par(mfrow = c(2,5))
plot(yearlyData[[1]]$Close,main = "",sub = "")
plot(yearlyData[[2]]$Close,main = "",sub = "")
plot(yearlyData[[3]]$Close,main = "",sub = "")
plot(yearlyData[[4]]$Close,main = "",sub = "")
plot(yearlyData[[5]]$Close,main = "",sub = "")

plot(yearlyData[[6]]$Close,main = "",sub = "")
plot(yearlyData[[7]]$Close,main = "",sub = "")
plot(yearlyData[[8]]$Close,main = "",sub = "")
plot(yearlyData[[9]]$Close,main = "",sub = "")
plot(yearlyData[[10]]$Close,main = "",sub = "")

Calculate the percentile of close with respect to 60 days previous prices.

subDays    <- lapply(d,FUN = function(x){return( (x-prevN):(x-2) ) })
subDays    <- as.Date(unlist(subDays))

dummyXts <- xts(order.by = subDays)
gb    <- gbXts[index(dummyXts)]

#Make High,Low,Close for each year
res <- to.yearly(gb)
colnames(res) <- colnames(gbXts)
res$percentile <- (res$Close - res$Low) / (res$High - res$Low)
res$percentile <- round(res$percentile, digits = 2)
cat(sprintf("Year = %d # percentile = %0.2f",year(index(res)),res$percentile),sep = "#  \n  ")

How many years the percentile is less than 60 (Meaning closer to the lows)? r nrow(res[res$percentile < 0.6,])
How many years the percentile is less than 50 (Meaning closer to the lows)? r nrow(res[res$percentile < 0.5,])
Total number of years : r nrow(res)

Compare with international prices

#Since international markets revolve mostly around US, the US price effect of today (say 02nd) is reflected on (03rd) of India.
#So, lag the prices by 1 day to compare apples to apples
gld   <- lag.xts(GLD[index(dummyXts)])
a <- merge.xts(gld$Close,gb$Close,join = "inner")
a <- na.trim(a)
colnames(a) <- c("intlClose","indiaClose")
cor(a$intlClose,a$indiaClose)

How many times the prise rise in US reflected the price rise in India ? r cor(a$intlClose,a$indiaClose)

60 Days Conclusion

This concludes that almost r nrow(res[res$percentile < 0.6,]) * 10 percent of times the diwali prices drifted lower than 60 percentile and r cor(a$intlClose,a$indiaClose)*100 percent times is closely correlated with international prices.

30 Days Evaluation

==================

prevN <- 30
yearlyData <- lapply(d,
                     FUN = function(x){
                       tmp <- (x-prevN):(x-2)
                       gbXts[as.Date(tmp)]
                       }
                     )

Graph the 30 days previous prices

par(mfrow = c(2,5))
plot(yearlyData[[1]]$Close,main = "",sub = "")
plot(yearlyData[[2]]$Close,main = "",sub = "")
plot(yearlyData[[3]]$Close,main = "",sub = "")
plot(yearlyData[[4]]$Close,main = "",sub = "")
plot(yearlyData[[5]]$Close,main = "",sub = "")

plot(yearlyData[[6]]$Close,main = "",sub = "")
plot(yearlyData[[7]]$Close,main = "",sub = "")
plot(yearlyData[[8]]$Close,main = "",sub = "")
plot(yearlyData[[9]]$Close,main = "",sub = "")
plot(yearlyData[[10]]$Close,main = "",sub = "")

Calculate the percentile of close with respect to 30 days previous prices.

subDays    <- lapply(d,FUN = function(x){return( (x-prevN):(x-2) ) })
subDays    <- as.Date(unlist(subDays))

dummyXts <- xts(order.by = subDays)
gb    <- gbXts[index(dummyXts)]

#Make High,Low,Close for each year
res <- to.yearly(gb)
colnames(res) <- colnames(gbXts)
res$percentile <- (res$Close - res$Low) / (res$High - res$Low)
res$percentile <- round(res$percentile, digits = 2)
cat(sprintf("Year = %d # percentile = %0.2f",year(index(res)),res$percentile),sep = "#  \n  ")

How many years the percentile is less than 60 (Meaning closer to the lows)? r nrow(res[res$percentile < 0.6,])
How many years the percentile is less than 50 (Meaning closer to the lows)? r nrow(res[res$percentile < 0.5,])
Total number of years : r nrow(res)

Compare with international prices

#Since international markets revolve mostly around US, the US price effect of today (say 02nd) is reflected on (03rd) of India.
#So, lag the prices by 1 day to compare apples to apples
gld   <- lag.xts(GLD[index(dummyXts)])
a <- merge.xts(gld$Close,gb$Close,join = "inner")
a <- na.trim(a)
colnames(a) <- c("intlClose","indiaClose")
cor(a$intlClose,a$indiaClose)

How many times the prise rise in US reflected the price rise in India ? r cor(a$intlClose,a$indiaClose)

30 Days Conclusion

This concludes that almost r nrow(res[res$percentile < 0.6,]) * 10 percent of times the diwali prices drifted lower than 60 percentile and r cor(a$intlClose,a$indiaClose)*100 percent times is closely correlated with international prices.

sivadevops1/playbook documentation built on May 31, 2019, 4:55 a.m.