Quantitative strategies on High Frequency Data

Posted 2021-02-05 javapythonc

Quantitative strategies on High
Frequency Data
Assessment of the sections (labs)
dr Piotr Wójcik
academic year 2018/2019
General information
In teams of at most 2 persons students will build and backtest different trading strategies
for 2 groups of assets. Please inform the lecturer about the team members by email
[email protected] (mailto:[email protected]) the latest by midnight 2018-
12-14.
The data is exactly the same for all teams and consists of intraday data of 1 minute
frequency (in xts format). The data covers real market data for the period of 2011-01 –
2012-03 (just trading hours 9:30-16:00 NY time) and is divided in 5 quarterly files.
For the purpose of strategy selection and parameters search students are initially given
the data just for 3 in-sample quarters. The remaining data for 2 out-of-sample quarters
will be delivered by the lecturer after the submission of presentations (see below).
General information
Quantitative strategies on High Frequency Data Page 1 of 7
file:///D:/FR%20video/warsaw/Quantitative%20Strategies%20on%20High%20Freque... 2018/12/30
Groups of assets
The groups of assets include:
1. Group 1 – a single asset:
ES – futures contract for S&P 500 index (transaction cost = 4$, 1 index point
= 50$).
2. Group 2 – three assets:
MSFT – Microsoft stocks (transaction cost = 0.2$)
AAPL – Apple stocks (transaction cost = 1$),
NQ – futures contract for NASDAQ index (transaction cost = 4$, 1 index point
= 25$).
CAUTION: The quarterly data files include all four assets –
please remember to treat ES (group 1) and MSFT , AAPL , NQ
(group 2) separately in your trading strategies.
Any combinations within group
allowed
Within each of the above groups of assets you can:
trade just a single asset, or
put assets together in pair(s) as spreads, or
trade each asset separately and treat them as a portfolio.
If trading more than one asset ( spread ), remember to include positive transaction
costs for each of them.
Trading sizes
Assume trading just with one unit of any security/spread, so the only positions available
are:
flat / neutral ( 0 ),
short ( -1 ),
long ( +1 ).
Different approaches, entry/exit
techniques
For each of the (groups of) assets please consider and compare at least 2 different types
of entry techniques (approaches), each with several combinations of parameters
(memories of moving statistics, multipliers, etc.).
As different approaches one may treat (each for the trend following or mean reverting
strategy) for example an entry/exit technique based on:
Quantitative strategies on High Frequency Data Page 2 of 7
file:///D:/FR%20video/warsaw/Quantitative%20Strategies%20on%20High%20Freque... 2018/12/30
? a single moving average/moving median/moving quantile,
two or more intersecting moving averages/moving medians/moving quantiles,
a single moving average/moving median/moving quantile and a selected volatility
measure (breakout models),
any other that comes to your mind.
Additional filtering
Additional filtering may be added (eg. in pair trading strategies):
based on correlation between two (or more) assets,
based on regression between two (or more) assets,
based on testing for cointegration between two (or more) assets,
based on testing for Granger causality between two (or more) assets,
any other that comes to your mind.
Common assumptions
Common assumptions for all strategies:
do not use in calculations the data from the first and last 10 minutes of the
session ( 9:31--9:40 and 15:51--16:00 ) – put missing values there,
do not hold positions overnight (exit all positions 15 minutes before the session
end, i.e. at 15:45 ),
do not trade within the first 20 minutes of stocks quotations ( 9:31--9:50 ), but DO
use the data for 9:41--9:50 in calculations of signal, volatility, etc.,
One may make additional assumptions, however they should be clearly explained and
justified, e.g. stop-loss condition, etc.
Selection of best strategy
CAUTION !!!!! As mentioned before, the data are divided in two
parts – in-sample quarters and out-of-sample quarters. At first
teams are provided just with the in-sample data to do a research
and select the best strategy for each group of assets
separately.
Exactly the same strategy (the same entry/exit technique and parameters) has to be
applied for a particular group of assets in each quarter.
For example if after research you find that for assets in group 1 the best strategy is a trend
following strategy based on the cross-over of two exponential moving averages – EMA60
and EMA10 – you should apply this particular strategy with the same parameters and all
other assumptions to every quarter of your data (first in-sample, then out-of-sample once
available) and report the results. The same in case of the second group of assets – the
best/optimal strategy may be different than in case of assets from group 1, but again – it
has to be consistently applied on all quarters of data.
Quantitative strategies on High Frequency Data Page 3 of 7
file:///D:/FR%20video/warsaw/Quantitative%20Strategies%20on%20High%20Freque... 2018/12/30
Selecting different best strategies (or just different parameters) for
the same group of assets in different quarters of the data is not
allowed.
Performance measures
For the selected best strategy for each group of assets aggregate the strategy P&Ls
to daily and based on daily results calculate the following measures (separately for each
quarter):
gross SR – Sharpe ratio based on gross daily P&L (without transaction costs,
denoted in monetary terms),
net SR – Sharpe ratio based on net daily P&L (with transaction costs included,
denoted in monetary terms),
gross cumP&L – cumulative profit and loss at the end of the investment period
(last value of the cumP&L series) without transaction costs, denoted in monetary
terms,
net cumP&L – cumulative profit and loss at the end of the investment period (last
value of the cumP&L series) with transaction costs included, denoted in monetary
terms,
av.ntrades – average daily number of trades.
and report them in a table at the end of the presentation and report.
Based on the above mentioned measures the final summary statistic will be calculated
for each quarter separately. The formula for the summary statistic is the following:
This promotes strategies which give high net Sharpe ratios and higher net pnl and strongly
awards for higher trading frequency.
Please add this statistic to the summary table and in addition use codes that will save
this table as a csv file.
The above mentioned summary statistic will be averaged separately for in-sample and
out-of-sample quarters and finally its weighted average ( 50\% for in-sample and 50\%
for out-of-sample results) will be used to rank the teams, divide them in quartile groups
and give points for strategy performance.
Points
In total 100 points can be collected, given for:
presentation in class prepared in RMarkdown including working R codes
( 20 pts ),
final written report prepared in RMarkdown including working R codes ( 40 pts ),
strategies performance ( 40 pts ) – ranking based on a summary statistic
described above, max. 20 pts. per each (group) of assets results:
stat = max(1, ln(1 + av. ntrades)) abs(netSR) sign(netSR)
abs(net. PnL) √
10
Quantitative strategies on High Frequency Data Page 4 of 7
file:///D:/FR%20video/warsaw/Quantitative%20Strategies%20on%20High%20Freque... 2018/12/30
? 20 if strategy performance in top quartile group (best),
15 if strategy performance in the 2nd quartile group (good),
10 if strategy performance in the 3rd quartile group (below average),
5 if strategy performance in the 4th quartile group (unlucky),
Presentations
The presentation prepared in RMarkdown has to be submitted by email to the lecturer
[email protected] (mailto:[email protected]) until midnight 2019-01-20
(presentation should be submitted both as the source *.Rmd file and also in the version
compiled to html , pdf or docx format). The R codes included in the Rmd file should
load the data from source files for each quarter, apply the BEST finally selected strategy
on the data for ALL quarters, calculate P&Ls and report the results in the desired form.
Do NOT include all the testing codes which you applied for
strategy selection, parameter search, etc. ONLY a simple code for
a FINALLY selected strategy i.e. with the selected set of best
performing parameters for each group of assets – check the
attached sample Rmd files prepared by the lecturer.
All teams will give presentations (10 minutes) informing about strategies considered and
their in-sample results. The presentations do not have to inform about all the details of
considered strategies.
Only teams that submit presentations in a desired format with
working R codes behind will obtain the out-of-sample data.
Teams which do not provide Rmd file with working R codes behind
which apply their best strategies will not be valued.
All presentations will take place on 2019-01-22 (labs time, 9:45-13:05 in room I).
Groups that do NOT present their results in class will get 0
points for presentation and for the out-of-sample
performance part.
Out-of-sample data
After all the presentations on 2019-01-23 the lecturer will provide the out-of-sample
data to enable verifying the strategy performance and finishing a final report. Having
prepared the report in R Markdown with working R codes behind will make your analysis of
the out-of-sample data very quick and on the other hand would allow the lecturer to verify
reported results and check if all assumptions are met.
Quantitative strategies on High Frequency Data Page 5 of 7
file:///D:/FR%20video/warsaw/Quantitative%20Strategies%20on%20High%20Freque... 2018/12/30
Final report
The final written report should be submitted by midnight 2019-01-29 (report should be
submitted both as the source *.Rmd file and also in the version compiled to html , pdf
or docx format). It should include a detailed explanation of the finally selected strategy
for each group of assets (approach, type and elements of the strategy, entry technique,
assumptions, parameter values, etc.) and also shortly explain the process of final strategy
selection. Measures of strategy performance should also be reported in a table (gross and
net SR, gross and net cum P&L, average daily number of trades) together with at least
one figure showing gross and net cumulative P&L of the strategy (based on daily
aggregated data).
Students who do not submit their final report before the deadline
will not be allowed to take the final written exam in winter
session.
Important dates again
2018-12-14 by 23:59 – submission of information about the team members
2019-01-20 by 23:59 – submission of presentations of in-sample results in R
Markdown with working R codes behind,
2019-01-22 – in class presentations of in-sample results – after that obtaining outof-sample
data.
2019-01-29 by 23:59 – final report submission
Each submission should be done via email to [email protected]
(mailto:[email protected]) before midnight of the deadline day if not stated
otherwise.
GOOD LUCK !!!!!!!!!!!!!!

因为专业，所以值得信赖。如有需要，请加QQ：99515681 或邮箱：[email protected] 

微信：codinghelp