Friday 31 July 2020

Currency Strength Candlestick Chart

In my previous posts on currency strength indices I have always visualised the indicator(s) as a line chart, e.g. here. However, after some deep thought, I have now created a way to visualise this as a candlestick chart using Octave's candle function, which, by the way, was written by me. Creating the candlestick body of a currency strength index was quite straight forward - just use the previous currency strength value as the bar's open and the current currency strength value as the close. A simple plot of this, with an overlaid currency strength index line chart, is

Of course the problem with this rendering is that there are no candlestick wicks.

My solution to create the wicks is showcased by the following code snippets
retval_high_wicks( ii , 6 ) = log( str2double( S.candles{ ii }.mid.h ) / max( str2double( S.candles{ ii }.mid.o ) , str2double( S.candles{ ii }.mid.c ) ) ) ;
retval_low_wicks( ii , 6 ) = log( str2double( S.candles{ ii }.mid.l ) / min( str2double( S.candles{ ii }.mid.o ) , str2double( S.candles{ ii }.mid.c ) ) ) ;
and
[ ii , ~ , v ] = find( [ retval_high_wicks( : , [32 33 34 35 36 37].+5 ) , -1.*retval_low_wicks( : , [5 20 28].+5 ) ] ) ;
new_index_high_wicks( : , 13 ) = accumarray( ii , v , [] , @mean ) ;
[ ii , ~ , v ] = find( [ retval_low_wicks( : , [32 33 34 35 36 37].+5 ) , -1.*retval_high_wicks( : , [5 20 28].+5 ) ] ) ;
new_index_low_wicks( : , 13 ) = accumarray( ii , v , [] , @mean ) ;
The first snippet shows an additional bit of code to the code here to record the log values of highs (lows) over (under) the candlestick bodies of all relevant currencies used in creating the currency strength indices.

The second snippet shows how the wicks are created, namely by taking the mean log values of high (low) wicks indexed by e.g.
[32 33 34 35 36 37].+5 and [5 20 28].+5
columns of downloaded forex crosses.

The reasoning behind this is as follows: take, for example, the EUR_USD forex pair - the upper wicks of these bars are recorded as upper wicks for the EUR index candles and as lower wicks for the USD index candles, reflecting the fact that upper wicks in EUR_USD can be viewed as intrabar EUR strength pushing to new highs or, alternatively, USD index candle's weakness pushing to new lows which, because the USD is the quote currency of the pair, also leads to new highs in the cross. A similar, reversed logic applies to the low wicks of the cross.

Below are charts of currency strength index candles created according to this methodology
The upper pane shows GBP currency strength index candles and the lower pane the same for USD. This is basically price action for Thursday, 30th July, 2020. The green vertical lines are the London and New York opens respectively, the red vertical line is the London close and the charts end at more or less the New York close. Bars prior to the London open are obviously the overnight Asian session.

My contemporaneous volume profile chart is the upper right pane below
 
From these charts it is easy to discern that the upward movement of GBP_USD during the main London session was due to GBP strength, whilst after the London close the continued upward movement of GBP_USD was due to USD weakness.

However, the point of this blog post was not to pass commentary on FX price movements, but to illustrate a methodology of creating candlestick charts for currency strength indices.

Enjoy!

Wednesday 15 July 2020

Forex Intraday Seasonality

Over the last week or so I have been reading about/investigating this post's title matter. Some quotes from various papers' abstracts on the matter are:
  • "We provide empirical evidence that the unique signature of the FX market seasonality is indeed due to the different time zones market participants operate from. However, once normalised using our custom-designed procedure, we observe a pattern akin to equity markets. Thus, we have revealed an important FX market property that has not been reported before." - Phd. paper - April 2013
  • "Using 10 years of high‐frequency foreign exchange data, we present evidence of time‐of‐day effects in foreign exchange returns through a significant tendency for currencies to depreciate during local trading hours. We confirm this pattern across a range of currencies and find that, in the case of EUR/USD, it can form a simple, profitable trading strategy" - Paper date - November 2010 - emphasis is mine
  • "This paper examines the intraday seasonality of transacted limit and market orders in the DEM/USD foreign exchange market. Empirical analysis of completed transactions data based on the Dealing 2000-2 electronic inter-dealer broking system indicates significant evidence of intraday seasonality in returns and return volatilities under usual market conditions. Moreover, analysis of realised tail outcomes supports seasonality for extraordinary market conditions across the trading day." - Paper date - May 2007
  • "In this article, we search for the evidence of intraweek and intraday anomalies on the spot foreign exchange (FOREX) market. Having in mind the international scope of this market ... We find that intraday and interaction between day and hour anomalies are present in trading EUR/USD on the spot FOREX market over the period of 10 years" - Paper date - 2014
  • "We find that the underpinnings for the time-varying pattern of the probability of informed trading are rooted in the strategic arrival of informed traders on a particular hour-of-day, day-of-week, and geographic location (market)." - Paper date - April 2008
In addition to this there seem to be numerous blogs, articles online etc. which also suggest that forex seasonality is a real phenomenon, so I thought I'd have a quick look into it myself.

Rather than do a full, statistical analysis I have used the following Octave function
clear all ;
data = dlmread( '/home/path/to/hourly_currency_index_g_mults' ) ;
## aud_x = x( 1)  ; cad_x = x( 2 ) ; chf_x = x( 3 ) ; eur_x = x( 4 ) ; gbp_x = x( 5 ) ; hkd_x = x( 6 ) ;
## jpy_x = x( 7 ) ; nzd_x = x( 8 ) ; sgd_x = x( 9 ) ; usd_x = x( 10 ) ; ## plus 6 for ix to account for date cols
## first 6 cols are YYYY MM DD HH-GMT HH-BST HH-EST
logged_data = data ; logged_data( : , 7 : end ) = log( logged_data( : , 7 : end ) ) ;

## get the days. The days of the week are numbered 1–7 with the first day being Sunday.
days_num = weekday( datenum( [ data(:,1) , data(:,2) , data(:,3) , data(:,5) ] ) ) ; ## BST time

start = input( 'Do you want to enter start date? Y or N ' , 's' ) ;
if ( strcmp( tolower( start ) , 'y' ) )
 year_start = input( 'Enter year YYYY:  ' ) ;
 month_start = input( 'Enter month MM:  ' ) ;
 day_start = input( 'Enter day date:  ' ) ;
 delete_ix = find( (logged_data(:,1)==year_start) .* (logged_data(:,2)==month_start) .* (logged_data(:,3)==day_start) ) ;
 
 if ( !isempty( delete_ix ) )
 logged_data( 1 : delete_ix , : ) = [] ; days_num( 1 : delete_ix , : ) = [] ;
 else
 disp( 'Invalid start date, so charts will show all data.' ) ;
 endif

endif

## create individual day indices
monday_indices = [ ( 0 : 1 : 23 )' , zeros( 24 , 10 ) ] ;
tuesday_indices = monday_indices ;
wednesday_indices = monday_indices ;
thursday_indices = monday_indices ;
friday_indices = monday_indices ;
alldays_indices = monday_indices ;

running_denom = zeros( 24 , 10 ) ;

for jj = 0 : 23
ix = find( ( days_num == 2 ) .* ( logged_data( : , 5 ) == jj ) ) ;
running_denom( jj + 1 , : ) = running_denom( jj + 1 , : ) + numel( ix ) ;
monday_indices( jj + 1 , 2 : end ) = sum( logged_data( ix , 7 : end ) , 1 ) ./ numel( ix ) ;
alldays_indices( jj + 1 , 2 : end ) = sum( logged_data( ix , 7 : end ) , 1 ) ;
endfor

for jj = 0 : 23
ix = find( ( days_num == 3 ) .* ( logged_data( : , 5 ) == jj ) ) ;
running_denom( jj + 1 , : ) = running_denom( jj + 1 , : ) + numel( ix ) ;
tuesday_indices( jj + 1 , 2 : end ) = sum( logged_data( ix , 7 : end ) , 1 ) ./ numel( ix ) ;
alldays_indices( jj + 1 , 2 : end ) = alldays_indices( jj + 1 , 2 : end ) .+ sum( logged_data( ix , 7 : end ) , 1 ) ;
endfor

for jj = 0 : 23
ix = find( ( days_num == 4 ) .* ( logged_data( : , 5 ) == jj ) ) ;
running_denom( jj + 1 , : ) = running_denom( jj + 1 , : ) + numel( ix ) ;
wednesday_indices( jj + 1 , 2 : end ) = sum( logged_data( ix , 7 : end ) , 1 ) ./ numel( ix ) ;
alldays_indices( jj + 1 , 2 : end ) = alldays_indices( jj + 1 , 2 : end ) .+ sum( logged_data( ix , 7 : end ) , 1 ) ;
endfor

for jj = 0 : 23
ix = find( ( days_num == 5 ) .* ( logged_data( : , 5 ) == jj ) ) ;
running_denom( jj + 1 , : ) = running_denom( jj + 1 , : ) + numel( ix ) ;
thursday_indices( jj + 1 , 2 : end ) = sum( logged_data( ix , 7 : end ) , 1 ) ./ numel( ix ) ;
alldays_indices( jj + 1 , 2 : end ) = alldays_indices( jj + 1 , 2 : end ) .+ sum( logged_data( ix , 7 : end ) , 1 ) ;
endfor

for jj = 0 : 20 ## market closes at 17:00 EST
ix = find( ( days_num == 6 ) .* ( logged_data( : , 5 ) == jj ) ) ;
running_denom( jj + 1 , : ) = running_denom( jj + 1 , : ) + numel( ix ) ;
friday_indices( jj + 1 , 2 : end ) = sum( logged_data( ix , 7 : end ) , 1 ) ./ numel( ix ) ;
alldays_indices( jj + 1 , 2 : end ) = alldays_indices( jj + 1 , 2 : end ) .+ sum( logged_data( ix , 7 : end ) , 1 ) ;
endfor

alldays_indices( : , 2 : end ) = alldays_indices( : , 2 : end ) ./ running_denom ;

monday_indices( : , 2 : end ) = cumsum( monday_indices( : , 2 : end ) ) ;
tuesday_indices( : , 2 : end ) = cumsum( tuesday_indices( : , 2 : end ) ) ;
wednesday_indices( : , 2 : end ) = cumsum( wednesday_indices( : , 2 : end ) ) ;
thursday_indices( : , 2 : end ) = cumsum( thursday_indices( : , 2 : end ) ) ;
friday_indices( : , 2 : end ) = cumsum( friday_indices( : , 2 : end ) ) ;
alldays_indices( : , 2 : end ) = cumsum( alldays_indices( : , 2 : end ) ) ;

if ( ishandle(1) )
 clf(1) ;
endif
figure( 1 ) ;
h1 = axes( 'position' , [ 0.03 , 0.54 , 0.30 , 0.43 ] ) ; plot( monday_indices(:,3) , 'k' , 'linewidth' , 2 , ...
monday_indices(:,4) , 'c' , 'linewidth' , 2 , ...
monday_indices(:,5) , 'b' , 'linewidth' , 2 , ...
monday_indices(:,6) , 'r' , 'linewidth' , 2 , ...
monday_indices(:,11) , 'g' , 'linewidth' , 2 ) ; xlim([0 23]) ; grid minor on ; title( 'MONDAY' ) ;
legend( 'CAD' , 'CHF' , 'EUR' , 'GBP' , 'USD' , 'location' , 'north' , 'orientation' , 'horizontal' ) ;
vline( 7 , 'r' ) ; vline( 12 , 'g' ) ;

h2 = axes( 'position' , [ 0.36 , 0.54 , 0.30 , 0.43 ] ) ; plot( tuesday_indices(:,3) , 'k' , 'linewidth' , 2 , ...
tuesday_indices(:,4) , 'c' , 'linewidth' , 2 , ...
tuesday_indices(:,5) , 'b' , 'linewidth' , 2 , ...
tuesday_indices(:,6) , 'r' , 'linewidth' , 2 , ...
tuesday_indices(:,11) , 'g' , 'linewidth' , 2 ) ; xlim([0 23]) ; grid minor on ; title( 'TUESDAY' ) ;
legend( 'CAD' , 'CHF' , 'EUR' , 'GBP' , 'USD' , 'location' , 'north' , 'orientation' , 'horizontal' ) ;
vline( 7 , 'r' ) ; vline( 12 , 'g' ) ;

h3 = axes( 'position' , [ 0.69 , 0.54 , 0.30 , 0.43 ] ) ; plot( wednesday_indices(:,3) , 'k' , 'linewidth' , 2 , ...
wednesday_indices(:,4) , 'c' , 'linewidth' , 2 , ...
wednesday_indices(:,5) , 'b' , 'linewidth' , 2 , ...
wednesday_indices(:,6) , 'r' , 'linewidth' , 2 , ...
wednesday_indices(:,11) , 'g' , 'linewidth' , 2 ) ; xlim([0 23]) ; grid minor on ; title( 'WEDNESDAY' ) ;
legend( 'CAD' , 'CHF' , 'EUR' , 'GBP' , 'USD' , 'location' , 'north' , 'orientation' , 'horizontal' ) ;
vline( 7 , 'r' ) ; vline( 12 , 'g' ) ;

h4 = axes( 'position' , [ 0.03 , 0.04 , 0.30 , 0.43 ] ) ; plot( thursday_indices(:,3) , 'k' , 'linewidth' , 2 , ...
thursday_indices(:,4) , 'c' , 'linewidth' , 2 , ...
thursday_indices(:,5) , 'b' , 'linewidth' , 2 , ...
thursday_indices(:,6) , 'r' , 'linewidth' , 2 , ...
thursday_indices(:,11) , 'g' , 'linewidth' , 2 ) ; xlim([0 23]) ; grid minor on ; title( 'THURSDAY' ) ;
legend( 'CAD' , 'CHF' , 'EUR' , 'GBP' , 'USD' , 'location' , 'north' , 'orientation' , 'horizontal' ) ;
vline( 7 , 'r' ) ; vline( 12 , 'g' ) ;

h5 = axes( 'position' , [ 0.36 , 0.04 , 0.30 , 0.43 ] ) ; plot( friday_indices(:,3) , 'k' , 'linewidth' , 2 , ...
friday_indices(:,4) , 'c' , 'linewidth' , 2 , ...
friday_indices(:,5) , 'b' , 'linewidth' , 2 , ...
friday_indices(:,6) , 'r' , 'linewidth' , 2 , ...
friday_indices(:,11) , 'g' , 'linewidth' , 2 ) ; xlim([0 21]) ; grid minor on ; title( 'FRIDAY' ) ;
legend( 'CAD' , 'CHF' , 'EUR' , 'GBP' , 'USD' , 'location' , 'north' , 'orientation' , 'horizontal' ) ;
vline( 7 , 'r' ) ; vline( 12 , 'g' ) ;

h6 = axes( 'position' , [ 0.69 , 0.04 , 0.30 , 0.43 ] ) ; plot( alldays_indices(:,3) , 'k' , 'linewidth' , 2 , ...
alldays_indices(:,4) , 'c' , 'linewidth' , 2 , ...
alldays_indices(:,5) , 'b' , 'linewidth' , 2 , ...
alldays_indices(:,6) , 'r' , 'linewidth' , 2 , ...
alldays_indices(:,11) , 'g' , 'linewidth' , 2 ) ; xlim([0 23]) ; grid minor on ; title( 'ALL DAYS COMBINED' ) ;
legend( 'CAD' , 'CHF' , 'EUR' , 'GBP' , 'USD' , 'location' , 'north' , 'orientation' , 'horizontal' ) ;
vline( 7 , 'r' ) ; vline( 12 , 'g' ) ;

if ( ishandle(2) )
 clf(2) ;
endif
figure( 2 ) ;
h1 = axes( 'position' , [ 0.03 , 0.54 , 0.30 , 0.43 ] ) ; plot( monday_indices(:,2) , 'k' , 'linewidth' , 2 , ...
monday_indices(:,7) , 'c' , 'linewidth' , 2 , ...
monday_indices(:,8) , 'b' , 'linewidth' , 2 , ...
monday_indices(:,9) , 'r' , 'linewidth' , 2 , ...
monday_indices(:,10) , 'g' , 'linewidth' , 2 ) ; xlim([0 23]) ; grid minor on ; title( 'MONDAY' ) ;
legend( 'AUD' , 'HKD' , 'JPY' , 'NZD' , 'SGD' , 'location' , 'north' , 'orientation' , 'horizontal' ) ;
vline( 7 , 'r' ) ; vline( 12 , 'g' ) ;

h2 = axes( 'position' , [ 0.36 , 0.54 , 0.30 , 0.43 ] ) ; plot( tuesday_indices(:,2) , 'k' , 'linewidth' , 2 , ...
tuesday_indices(:,7) , 'c' , 'linewidth' , 2 , ...
tuesday_indices(:,8) , 'b' , 'linewidth' , 2 , ...
tuesday_indices(:,9) , 'r' , 'linewidth' , 2 , ...
tuesday_indices(:,10) , 'g' , 'linewidth' , 2 ) ; xlim([0 23]) ; grid minor on ; title( 'TUESDAY' ) ;
legend( 'AUD' , 'HKD' , 'JPY' , 'NZD' , 'SGD' , 'location' , 'north' , 'orientation' , 'horizontal' ) ;
vline( 7 , 'r' ) ; vline( 12 , 'g' ) ;

h3 = axes( 'position' , [ 0.69 , 0.54 , 0.30 , 0.43 ] ) ; plot( wednesday_indices(:,2) , 'k' , 'linewidth' , 2 , ...
wednesday_indices(:,7) , 'c' , 'linewidth' , 2 , ...
wednesday_indices(:,8) , 'b' , 'linewidth' , 2 , ...
wednesday_indices(:,9) , 'r' , 'linewidth' , 2 , ...
wednesday_indices(:,10) , 'g' , 'linewidth' , 2 ) ; xlim([0 23]) ; grid minor on ; title( 'WEDNESDAY' ) ;
legend( 'AUD' , 'HKD' , 'JPY' , 'NZD' , 'SGD' , 'location' , 'north' , 'orientation' , 'horizontal' ) ;
vline( 7 , 'r' ) ; vline( 12 , 'g' ) ;

h4 = axes( 'position' , [ 0.03 , 0.04 , 0.30 , 0.43 ] ) ; plot( thursday_indices(:,2) , 'k' , 'linewidth' , 2 , ...
thursday_indices(:,7) , 'c' , 'linewidth' , 2 , ...
thursday_indices(:,8) , 'b' , 'linewidth' , 2 , ...
thursday_indices(:,9) , 'r' , 'linewidth' , 2 , ...
thursday_indices(:,10) , 'g' , 'linewidth' , 2 ) ; xlim([0 23]) ; grid minor on ; title( 'THURSDAY' ) ;
legend( 'AUD' , 'HKD' , 'JPY' , 'NZD' , 'SGD' , 'location' , 'north' , 'orientation' , 'horizontal' ) ;
vline( 7 , 'r' ) ; vline( 12 , 'g' ) ;

h5 = axes( 'position' , [ 0.36 , 0.04 , 0.30 , 0.43 ] ) ; plot( friday_indices(:,2) , 'k' , 'linewidth' , 2 , ...
friday_indices(:,7) , 'c' , 'linewidth' , 2 , ...
friday_indices(:,8) , 'b' , 'linewidth' , 2 , ...
friday_indices(:,9) , 'r' , 'linewidth' , 2 , ...
friday_indices(:,10) , 'g' , 'linewidth' , 2 ) ; xlim([0 21]) ; grid minor on ; title( 'FRIDAY' ) ;
legend( 'AUD' , 'HKD' , 'JPY' , 'NZD' , 'SGD' , 'location' , 'north' , 'orientation' , 'horizontal' ) ;
vline( 7 , 'r' ) ; vline( 12 , 'g' ) ;

h6 = axes( 'position' , [ 0.69 , 0.04 , 0.30 , 0.43 ] ) ; plot( alldays_indices(:,2) , 'k' , 'linewidth' , 2 , ...
alldays_indices(:,7) , 'c' , 'linewidth' , 2 , ...
alldays_indices(:,8) , 'b' , 'linewidth' , 2 , ...
alldays_indices(:,9) , 'r' , 'linewidth' , 2 , ...
alldays_indices(:,10) , 'g' , 'linewidth' , 2 ) ; xlim([0 23]) ; grid minor on ; title( 'ALL DAYS COMBINED' ) ;
legend( 'AUD' , 'HKD' , 'JPY' , 'NZD' , 'SGD' , 'location' , 'north' , 'orientation' , 'horizontal' ) ;
vline( 7 , 'r' ) ; vline( 12 , 'g' ) ;
to conduct a quick visual analysis. This builds upon my recent work on fx pairs via oanda api and currency strength, and uses hourly data since June 2012.

This produces 24 hour seasonality charts of CAD, CHF, EUR, GBP and USD, i.e. the European and North American currencies.
The x-axis is in British Summer Time (BST) hours, the vertical red and green lines indicate 7:00am opens in London and New York respectively, all charts end at 17:00 New York (EST) time and the y-axis is hourly log returns. The individual currency seasonality lines are the cummulative cross-sectional means at BST 00, BST 01 ... etc. per weekday and all days combined (see subchart titles). BST Sunday evenings' returns prior to Monday trading are not included.

A similar chart for the Asian time zone currencies of AUD, HKD, JPY, NZD and SGD is also produced.
The function allows charts with a user selected data begin date to be plotted, but the illustrations above use all data available to me, i.e. hourly data since 2012.

It seems to me that, as indicated in my highlighting above, there is definite intraday forex seasonalilty in play. However, readers should be cautioned that the above is only a general tendency based on the last 9 years or so of hourly data. A more recent "data snapshot" of only data since the beginning of 2020 can tell a slightly different story:
look at GBP (red line) on Tuesdays and Wednesdays, for example. As always with stuff one reads online, even the extremely high quality stuff on this blog 😊, Caveat emptor.

Friday 10 July 2020

Time Warp Edit Distance as a Loss Function

Some time ago I posted about the time warp edit distance (twed) and recently I've revisited this as a possible loss function. The basic idea is to use this during algorithm training to match a "perfect equity curve" rather than use the usual loss measurements such as mean squared error for regression or cross entropy for classification.

As a proof of concept I've been playing around with this Octave script
clear all ;
1 ;

function J = twed_loss( x )
 global price ;
 global returns ;
 global perfect_equity_curve ; 
 fast_ma = sma( price , round( x( 1 ) ) ) ;
 slow_ma = sma( price , round( x( 2 ) ) ) ;
 test_position_vector = sign( fast_ma .- slow_ma ) ;
 test_position_vector = shift( test_position_vector , 2 ) ;
 test_position_vector( 1 : 3 ) = 0 ;
 test_equity_curve = cumsum( returns .* test_position_vector ) ;
 x_axis = ( 1 : numel( perfect_equity_curve ) )' ;
 J = twed( perfect_equity_curve , x_axis , test_equity_curve , x_axis , 1 , 0.001 ) ;
endfunction

## create a perfectly predictable price
global price = sinewave( 200 , 20 )' .+ 5 ;
global returns = [ 0 ; diff( price ) ] ;
perfect_position_vector = sign( returns ) ;
perfect_position_vector = shift( perfect_position_vector , 2 ) ; 
perfect_position_vector( 1 : 3 ) = 0 ;
global perfect_equity_curve = cumsum( returns .* perfect_position_vector ) ;

## now do the Baysian training
## set up the parameters for bayesopt
params.n_iterations = 300 ; ## 190
params.n_init_samples = 10 ; ## 10
params.n_iter_relearn = 1 ; ## Number of iterations between re-learning kernel parameters. That is, kernel learning ocur 1 out of n_iter_relearn iterations. 
## Ideally, the best precision is obtained when the kernel parameters are learned every iteration (n_iter_relearn=1). 
## However, this learning part is computationally expensive and implies a higher cost per iteration. If n_iter_relearn=0, then there is no relearning. [Default 50]
params.crit_name = 'cEI' ;
params.surr_name = 'sStudentTProcessNIG' ;
params.noise = 1e-6 ;
params.kernel_name = 'kMaternARD5' ;
params.kernel_hp_mean = [ 1 ] ;
params.kernel_hp_std = [ 10 ] ;
params.verbose_level = 0 ; ## Negative -> Error -> stdout 0 -> Warning -> stdout 1 -> Info -> stdout 2 -> Debug -> stdout
                           ## 3 -> Warning -> log file 4 -> Info -> log file 5 -> Debug -> log file 5 -> Error -> log file
params.load_save_flag = 0 ; ## 1-Load data, 2-Save data, 3-Load and append data. Other values, no file saving or restore [Default 0]
params.log_filename = '/home/dekalog/Documents/octave/twed/bayeopt.log' ; % Name/path of the log file 
## (if applicable, verbose_level>=3) [Default "bayesopt.log"]
params.load_filename = '/home/dekalog/Documents/octave/twed/bayeopt.log' ;
params.save_filename = '/home/dekalog/Documents/octave/twed/bayeopt.log' ;

lb = [ 2 3 ] ;
## upper bounds
ub = [ 30 30 ] ;
nDimensions = length( lb ) ;
[ xmin , fmin ] = bayesoptcont( 'twed_loss' , nDimensions , params , lb , ub ) ;
round( xmin )
fmin

fast_ma = sma( price , round( xmin(1) ) ) ;
slow_ma = sma( price , round( xmin(2) ) ) ;
figure(1) ; plot(price,'k','linewidth',2,fast_ma,'r','linewidth',2,slow_ma,'b','linewidth',2 ) ;
title( 'PRICE AND MA CROSSOVER SIGNALS' ) ; legend( 'PRICE' , 'FAST MA' , 'SLOW MA' ) ;
test_position_vector = sign( fast_ma .- slow_ma ) ;
test_position_vector = shift( test_position_vector , 2 ) ;
test_position_vector( 1 : 3 ) = 0 ;
test_equity_curve = cumsum( returns .* test_position_vector ) ; 
figure(2) ; plot(perfect_equity_curve,'b','linewidth',2,test_equity_curve,'r','linewidth',2) ;
title( 'EQUITY CURVES' ) ; legend( 'PERFECT EQUITY CURVE' , 'TEST EQUITY CURVE' ) ;
which produces plots such as this
and this,
which both show a fast and slow moving average crossover system on sine wave "price" of period 20, optimised to match equity curves such as below via the twed loss.
What is interesting about this is that the moving average lengths usually converge to more or less the expected theoretical optimum, with the required change of sign of signal, where the crossovers indicate peaks and troughs in price and hence perfect entry and exit signals.

However, sometimes the solution looks like this,
which is an 11 period fast moving average and a 10 period slow one, quite a contrarian solution compared to the theoretical optimum, but actually giving a lower twed loss.

I quite like the idea of optimising for what we actually care about, i.e. the equity curve, whilst at the same time possibly uncovering unique solutions. It seems that the twed loss shows promise.

More in due course.