In the culminating project, you will develop new trading strategies, evaluate them using the tools learned in the course, integrate them with the existing portfolio and also develop a plan to start a hedge fund.

Loading...

来自 Indian School of Business 的课程

Design your own trading strategy – Culminating Project

28 个评分

At Coursera, you will find the best lectures in the world. Here are some of our personalized recommendations for you

In the culminating project, you will develop new trading strategies, evaluate them using the tools learned in the course, integrate them with the existing portfolio and also develop a plan to start a hedge fund.

从本节课中

Week 2 - Strategy

In this module you will go through the paper based on
'Pairs Trading' strategy and understand the details of the strategy completely. You will also identify the sector and related sector on which you want to base your strategy.

- Ramabhadran ThirumalaiAssistant Professor

Indian School of Business

The next part is our empirical strategy.

The empirical strategy will tell you whether the statistical methods, tools and

techniques the authors use to prove that

whatever they're trying to establish actually happens in practice.

So this part is highly technical.

Let me warn you unless you are trained in econometrics,

unless you're trained and have some background in mathematics and

statistics it may be difficult to interpret these things.

For example you may suddenly find a sentence that we use regression

discontinuity technique.

Now what is this regression discontinuity?

So if you try understanding it, it may take a lot of time,

which may not be required purely from a trading point of view.

So this empirical strategy section can be safely ignored

from a purely trading point of view.

Of course, it's great to learn, but not required from a trading point of view.

Coming back, the next section is important, this is about results.

Here the author show what has been the performance of these strategy.

This is very, very important.

Now, here is the difference between an academic exercise and

a data mining exercise.

Now what happens in a data mining exercise, the purpose of each test, here

they're in a pure data mining exercise, is to show that something actually works.

So authors there, people try like ten tests and

pick up the one that works, and say that this works.

If you have seen a typical technical analysis kind of

a show where people say that some particular pattern works.

Head and Shoulders, for example works.

If you don't know Head and Shoulders, nice.

You can think that it's a shampoo brand and forget it.

But then how do they prove?

They pick some instance where it works.

So they say that six months back on a particular store, Head and Shoulders

pattern emerged, and it worked, and that's the money it made, the strategy made.

That's why it's an amazing strategy.

Now again the Head and Shoulder pattern has emerged and

hence you are likely to make money if you follow this particular strategy.

This is a typical data mining kind of an approach.

The approach that academic papers take is not this.

The author attempts to reject a particular hypothesis,

continuously tries to show that a particular trading strategy does not work.

And if he fails to show that it does not work,

then he says, there is likelihood that this may work.

He will never ever conclude with certainty that a particular study will always work.

After a careful attempt to reject the possibility

that the strategy does not work then he concludes that probably this works.

It's a very different approach.

Now coming back to Head and Shoulders.

Now if you give it to a careful researcher with Head and Shoulders.

You know what he will do?

He will take data of 20 years, or

whatever, 30, 40, 50 years, that data is available for.

He will look at all instances where this pattern has emerged.

This is very important, he will not pick and choose.

Sometimes this worked, some 20 years back this shape emerged and

it worked, he will not do that.

What the careful researcher will do will look at all instances where this

shape emerged and then out of these instances, what has been the result?

How many times you made money?

How many times you lost money?

How many times nothing has happened?

Suppose these strategies worked let's say as on an average given you some 5% return.

Now the author compares this to what would have happened if you would have

just randomly traded, without the strategy being there.

Suppose you literally call some monkeys and

ask them to throw darts on a stock, what would have been your return?

You can assume market return that you'll get from

a passive trading strategy that you get from an index as a benchmark.

And they compare whether the strategy of buying whenever this

particular shape emerges outperforms this kind of passive investing.

The approach here is to attempt to reject something and despite your best efforts,

if we cannot reject, then you sort of say that maybe this is working.

So that's the approach these papers will take.

So, as I've said, this Head and Shoulders, they compare it with market returns.

And mind you, most of the time with these kind of strategies,

you will find that those strategies don't work in practice.

Just think of this way.

So if we could have just looked at some shape and made money,

everyone would have made money, right?

That's not that easy.

So, you should be very, very careful while looking at this self-selected examples.

Whenever you hear of some tip or

a trading strategy, you should always keep this in mind.

The approach that you're following is it trying to accept something or

reject something?

So from that point of view reading this result section is very important.

What this results section will do is that initially it

will show you the main result, the authors describe the main result.

And then they show you a series of tests which are performed where

the authors attempt to reject this particular hypothesis that they have.

And they fail to reject it.

That is why they claim that there is a paper,

there is an idea which can be exploited.

They also attempt to show that,

what if it's some luck because of which, this has worked?

They perform tests to sort of disapprove that.

And these are called robustness test.

Now what are these robustness tests?

Let me give you some examples.

It's quite possible that out of a 20-year period, you had of one or

two-year period which was extraordinary, let's take financial crisis.

You have a strategy where you short banks or any financial institution.

And you choose a sample, put it between say, 2005 and 2015, or 2012.

And you find that the strategy has worked on an average

and it has given you annualized, say 20% return, amazing.

But then, it's quite possible that all your returns

are concentrated between 2007 and 2009 or 2008.

What does this mean?

That means that this strategy has worked only during financial crisis.

Now if financial crisis were to persist forever, then it's fine, but

during normal times it may not work.

And more importantly, when the economy booms, this may go the other way,

that's more important.

Suppose you always strategy of going short on value stocks or distressed stocks.

When the economy is in a bad shape Or

when it is entering a crisis, entering a recession, this strategy may do very well.

But when the recovery starts, these are the very stocks which fell 90%,

95% during a crisis, may recover like three times, four times in no time.

Remember those who shorted City at very low levels,

they would have lost their shirts when the stocks recovered.

So, an average result over 20 years will not tell you all these things.

So the subsequent tests which authors perform, they try to control for

all these possible explanations.

So in technical terms, they have time fix defects,

which basically control for in part of time.

In other words, by employing these they are trying to show that these results

are not driven by a particular period, particular year, particular month.

Then they also show that these are not driven by few stocks.

And it's quite possible that your portfolio consists of 100 stocks, but

2 or 3 of them suddenly moved abnormally because of some reason, and

your entire portfolio makes positive return because of them.

Now this is not a replicable situation, because such two,

three stocks may not exist next time.

Please understand, our purpose is not just understanding the paper,

our purpose is to see whether we can utilize the strategy elsewhere.

So they use these kind of fixed effects like at firm level to rule out

that this is not driven by a particular firm or a particular group of firms.

What they also do is that they subdivide the sample,

as a further robustness they divided into some smaller time intervals and

they show that the strategy works in each time interval.

More importantly, each time interval, the economic situation may be very different.

They also show that this is not an impact of some alternative explanation.

Something else going on which caused this.

So all these tests are designed to rule out these kind of alternative

explanations, which could have caused these results And

hence made these results not replicable in another situation.

So, from that point of view, at least the first paper that you want to read,

try to see at least the purpose of these tables.

You may not understand exactly what the underlying econometrics is,

but at least, before this result section,

each subsection in that section, the author, in the first three,

four lines author clearly mentions why this particular test is done.

If you understand that and the fact that the paper is published,

has gone through peer review process, you can be reasonably sure that very,

very low hanging kind of objections the authors would have taken care of.

Does this mean that the paper is perfect?

Does this mean that there's no other alternative explanation possible?.

Does this mean that the strategy always works?.

No, I've told you right at the beginning.

We are not giving you lollipops here, low hanging fruit you just pick up and

you start making money from tomorrow, that's not what is going to happen.

But then, I can tell you with lot of confidence,

that the rigor that these papers have is unmatched.

This is far, far more reliable and

replicable than a result of a pure data mining exercise.

So once you go through this results section, so

how much attention you are to pay?

Now, let's recap a bit, I told you there is an abstract.

There is an introduction, then you have institutional background,

then you have data.

Then you have algorithm, trading algorithm, then you have empirical

strategy, then you have results, and finally there is conclusion.

Now before we get conclusion, in some papers,

you will also have a theory section.

Now, From our trading point of view you can just ignore it completely.

Now, it's a very important part, I'm not trying to belittle that section.

Basically, here the authors try to show that how their paper

contributes to the existing theory in economics, corporate finance,

accounting, whichever area that paper deals with,

but purely from a trading point of view, it's not required.

Now is it totally useless to understand theory?

No, not at all, but the point is this section

is extremely dense, highly mathematical.

You will have theorems, you will have proofs, so if you have a liking for

these things, please go ahead.

A positive side effect of understanding this section is that

this will help you to improvise.

Understanding theory will help you to improvise, but

then if you don't understand, it's fine.

You can still implement the strategy as it is.

So now coming back to the last part of your work, you will have conclusion.

So conclusion here again authors summarize the results, and show that how their

results contribute to the literature, what improvements can be done.

Now that is also,

some of the papers give you some clues as to what further work can be done.

You may just start from there and if it's a recent paper,

you can be sure that nobody would have done that also.

So from that point of view, this conclusion is important,

it's typically one, one and a half pages, sometimes less than that.

And with that the paper ends.

Then you have citations and then you have tables.

Now when you read, one thing that I forgot to tell you, when you read a data section,

as you read the data section, also see the relevant table.

First two, three tables will summarize data, data sources,

variable definitions and all that, so read them in parallel.

Also, when you read the results section, try to read,

because the way the paper is organized is first 30 pages, as I've told you,

first 30 pages of text, 25, 30 pages of text, citations, 3, 4 pages, then tables.

So text corresponding to our table and

the table are separated, they're not put together.

So when you read a particular section on results, or when a particular result,

I strongly encourage you to also look at tables simultaneously.

Then the understanding will be worth it.

When the author says, this strategy leads to 3% return on outlays,

just look at the table and verify for yourself, do you find that 3% somewhere?

That'll help you understanding better.

So you should read them together, so once in the table part,

then you will have pictures.

Picture basically help improve understanding and

that's how a paper is typically structured.

So now what we'll do, now that you understand how a paper looks like,

I encourage you to look at this.

If you see the slide it'll show you where this Piotroski paper is available.

I strongly encourage you to download this paper, maybe print it out or

use your device.

And just verify for yourself, where did you find these sections.

If you find or any other training strategies paper.

If you find anything which you cannot understand, you can write to us.

We will answer your questions.

And if you find any new section which I am not told.

There is no law.

This is not a legal format.

This is not a format determined by law.

You have, depending on the context of the paper, you may have other sections as well

or there will be papers that don't have any sections.

A total theory for example.

Most papers don't have theory, some papers may have it.

Some papers can may combine two or more sections.

For example, hypothesis and empirical strategy may be combined.

That's possible.

Now that you understand the structure of a paper I can assure

you that you will not get frightened by the technicalities involved in a paper.

Before proceeding further on this video, I strongly encourage

you to download this Petrosky paper or any other paper and just try to see yourself.

As I've said before, there could be change in the structure.

But more important for you is to understand where this trading strategy

section, or the algorithm section is, and start trying to replicate.

Another prerequisite for

this kind of course Is that you must be familiar with some kind of a statistical,

not a statistical package to analyze it.

It could be as straightforward as an Excel,

or if you know programming that's even better.

Not essential again, you can do lot of your work in Excel itself.

But I strongly encourage that you brush up your basics in any of these packets.

So that you can download data and start testing them.

That's very important.

Remember the key is to have this mindset that you're making

an attempt to reject the hypothesis and not to come out with something.

We ultimately are to put your hard earned money and

also your investors into this strategy.

So it's no fun in getting highly results on paper and

when you actually go and practice finding something else.

And that's possible if their testing is done in this kind of a,

with this attitude to find some results.

You always try to reject it and

I'm confident that on average you will do well.

So with this we will conclude this section on how to read a paper and

how do we go forward from here.

In the next part, I will pick up Petroski's paper and

take you through the abstract line by line.

Now the purpose of this of course I can't take you through

the entire paper line by line, that itself would be 12 hours or 15 hours.

And as I told you that's not needed.

Also what I've started take you line by line so

that you understand how to read the abstract and what does this paper do.

And then from there I'm going to jump straight away into the algorithm section.

And tell you exactly how Piotrosky Score works.

Once you know scoring, what I'll do is I'll explain how do you actually trade.

I'll also explain what are the kind of results that Piotrosky has obtained and

then I will also show you by taking of real example of a particular company

from India and how do you get these scores.

And you can only in other country what will happen is

where these items are described in balance sheet, ordering maybe different but

the broad Idea should work everywhere.

So that's the plan going forward.

So I heard you strongly,

I encourage you to just go through this Piotrowski paper, at least broadly,

before you start the next part on abstract and also the scoring of strategies.