Introduction to Zipline: A Trading Library for Python
quantinsti.com
quantinsti.com
I get asked quite a bit on how to start doing algorithmic trading and the first thing I always tell people is don't.
I think I've said this many times now but the number of people who come at it with the thinking "I'm a computer scientist. I'll just fire up R or python and apply some machine learning to the markets and watch the money roll in" is staggering.
I mean each day 100's of Phd's start with clean market data, more data sources than you could possibly think of and statistical back testing systems that have 1000's of man hours put into them, trying to find a way to make money.
After all of that if you really want to I wrote this in response to an Ask Hacker News a little while ago
https://news.ycombinator.com/item?id=11352562
TL/DR - focus on time periods greater than a day
- expect to lose money
- expect to take a year to figure out some edge in the market
- most decent trading strategies that a normal person can use come from economic/market insights first and technology second.
The site: https://www.quantstart.com/ is also decent at bringing you up to speed on the math you'll need to know though I believe that the material there oversells how easy it is to find a decent trading strategy.
A trader that is independent may still have advantages that prevent some random programmer from bootstrapping his/her way to also being an independent trader.
Just thinking out loud though, I'm not a trader of any sort.
The best for you to start learning - would be to get a start on how the market works - my suggestion would be - https://www.amazon.com/dp/B000THOD1G/
Case in point is the fluctuations in Warren Buffett's fund whenever Anne Hathaway is in the news.
http://ftalphaville.ft.com//2011/03/28/528481/for-the-bots-a...
There's actually a logical error here: if trading results were essentially random, a certain subset of traders (including day traders) would, at any given time, have profitable records. But you would not be able to derive (and, therefore, implement in code) any set of rules which would make automated trading profitable. You would just either be lucky, or not, as an algorithmic traders just as you would as any other kind of trader.
Now, I'm not saying that trading profits are random, but the existence of some profitable traders does not mean that there are rules you can deduce from their behavior that will guarantee profitable trading when implemented by someone else (either in an automated system or otherwise.)
It's relatively easy to have a profitable period on a strongly bull market. But it's irrelevant if any 'unexpected' (people call such events unexpected despite the fact that they tend to happen regularly over time) event such as 2008 crisis will completely wipe you out.
You're forgetting about the many that aren't profitable. You're also forgetting about the majority that are less profitable than the market average. The latter is probably the easiest to overlook. Yes, they are technically profitable, but so are packaged portfolios of stock held for long periods of time. If you can't be more profitable than someone who puts in zero effort, then whats the point of putting in more work?
Sure, it is possible to to program very niche behaviour, but we are nowhere near any sort of program that can act as general "day trader".
The stock markets are very far from a rational, perfect information game with simple rules.
Another thing that gets ignored is the the difference between a trading strategy and an execution algorithm.
don't need high frequency at all, just pretty average latency actually (for now anyway)
IB probably has this, not sure though. But those data costs are a huge deterrent for me!
anyway, I'm sure this is its own discussion
Checkout CDS and FI market data offerings: https://www.marketaxess.com/data/marketdata.php
But your idea is correct. CDS Spreads can be predictive of dramatic shocks in an equities price.
Yeah i get the promise: professionals do this, pay a premium for that
Until they don't because someone provides the data for free or next to nothing
In this town we call it disruption
Tradier has been providing equity and option tick data for free for years in a clean RESTful API with the capability of websockets and streaming. So thats the answer to your question of who.
Like I said, in this town we call it disruption.
We are unlikely to ever see CDS data for free since individuals cannot trade them as most retail investors are not Eligible Contract Participant. Some brokerages give away data to get people to trade with them so there its about customer acquisition.
The best way to make some money as a personal trader is to take advantage of the liquidity premium in one way or another. Because you're trading money in the 5 or 6 figures rather than the 7 or 8 figures, you can take advantage of smaller opportunities without thoroughly distorting the market with your own trades. These smaller trades usually require research and market insight rather than clever algorithms. These trades are usually on financial instruments that don't have a lot of easily accessible data to build an automated trade on top of.
This is very important. Most brokerage charge around $7 per trade, which makes high-volume trading very very expensive and prohibitive.
Robin Hood is an amazing alternative that charges nada for trades, and once they have an API[1], I think they'd be a great choice for small-time developers looking to do some (low-frequency) algorithmic trading.
[1] https://support.robinhood.com/hc/en-us/articles/210216823-Ro...
The next step down are firms that'll give you a platform including hardware and software for the "infrastructure" bits: often called an "algo container", but lots of brokers have an API you can use to avoid needing to write feed handlers, etc. eg. Pico.
Then there are heaps of providers who'll rent you servers, connectivity, rackspace, etc. You do all the software. Lucera is a trading-oriented "cloud" provider. OptionsIT or Fixnetics are infrastructure providers. Or you just go straight to the data centers -- any decent finance-oriented datacenter will have a POP for most of the venues, and you can just cross-connect.
No relationship with any of the firms named -- just examples off the top of my head.
The big issue isnt actually infa it's capital. If you want to trade big you need a million bucks or something on hold with the exchange.
How do you borrow a mil to park at an exchange? The guys mentioned above.
80% of profits seems a bit rich, though. If they're supplying top spec colo servers, a super fast network to connect them, and people to stare at it, I'd say something a bit lower (but still pretty chunky) would make more sense.
I suppose the key for them is they have a bunch of existing infrastructure that pays for itself, so any extra cream is good. Also they have capital sitting around, so why not?
The strategies have to be more than unusually profitable, though. I don't think I've seen a down day in the 3 months I've been sitting here, so your average trend following sharpe ~1.5 or so ain't gonna impress people.
That said, the same logic that holds for identifying profitable strategies within an institution holds for individuals: unless you have better gear, don't fish a crowded pond. As an individual, your small size is, in some regards, an advantage. Institutions routinely pass on strategies that don't have capacity (there's not enough liquidity to make the strategy's returns worthwhile relative to their trading level) or strategies that aren't quite up to their standards (but might be up to yours). There's also a huge class of strategies that have somewhat choppy but long run consistent returns. Traders and funds worried about MoM track records won't touch those.
Fees are the biggest barrier to entry. Institutions enjoy substantial discounts and an ability to amortize costs across a much wider base, making the strategy performance hurdle rate proportionally higher for smaller traders.
Interactive brokers offers low fee access to their platform (https://www.interactivebrokers.com/en/index.php?f=13869)
Quantopian (www.quantopian.com) gives you the ability to trade through robinhood with long trades at 0 commission. There has been some skepticism in the Q forums on how well robinhood's execution is (possibly the effect of a "you get what you pay for"-attitude).
Trading real capital with algorithms is more difficult than it sounds. Mentally if you're algorithm is doing stock selection as well as trade execution, you no longer understand what you own. During a period of extended drawdown, you have to be mentally tough enough to believe that your algorithm has what it takes to dig itself out.
In a sense, it is like the start-up game. You have to believe you're doing something well enough that in the end you'll be right (and not bankrupt).
Seriously, here's your competition: https://en.wikipedia.org/wiki/Renaissance_Technologies How deluded do you have to be to think you have an edge over that?
People like you stay out.
More to the point, just because a genius mathematician and code breaker started a hedge fund it doesn't at all push out any of the little guys. The market is so large he can't possibly be trading all instruments at once, and "scaling" is a problem for huge hedge funds. Especially ones that have to answer to their shareholders. Even though this is a huge fund, it is highly inflexible.
What you're saying is akin to "Google already does email and they're loaded with geniuses, what makes you think you have an edge over that?"
Or it will get done by people who have thought long and hard about how they are going to compete with the giants, rather than someone naïve sap.
The fact that the question is being asked does not imply that there are no good answers to it, but it is unlikely that someone who hasn't spent time considering how to improve on the incumbents is going to beat them.
Incorrect for the reasons pointed out above, they're trading size, you're likely not, it's a completely different game for them in a completely different field due to their liquidity needs. They are not your competition.
For example Taaffeite Capital Management (who gained some publicity for good returns on Brexit) are a < $10mm fund who use IB[1].
Sure, bigger funds have better market access, and it will always be impossible to implement a high frequency trading approach. But these platforms are about as good as a small player can get.
[1] http://www.afr.com/personal-finance/shares/artificial-intell...
I think this goes directly to the "if you use these platforms you are a fool" comment. It seems to me that some people who aren't fools use the platforms, and some who are, don't.
Also, this is HN. Pretty sure there is more than one MIT PhD reading this, and I know there is more than one doing original AI research. Bring this comment to mind: https://twitter.com/paulg/status/28911860225 (exact comment here: https://news.ycombinator.com/item?id=35079)
"if you use these platforms you are a fool"
I said if you use these platforms _and don't have a reason to think you've got an advantage_ you shouldn't be doing so.
Lun and his peers are an exception to this. If you want to make a second argument and say he shouldn't be trading go ahead (and I'll try to back you up), but my original point was about the 99.9% who aren't Lun and are clearly just fish.
If there's any downside, it's their data. You can't get that much of it historically (1 year max on the minute bars, far less on the second bars, and it takes forever to download it because of the throttling). It's also not...I'm going to say correct...historically. I mean, it is a correct record of what trades happened when. But it includes trades that you won't see when they happen, making it less useful if you're looking for a stream of events as they are happening. Also compared to a broker like Lightspeed (who I haven't tried) their data is expensive.
The software is a big mess, but it's usable. It's always in this weird state where it's 50-75% of the way to being completely awesome, but there's just a few things missing that stop it from being so. Also if I have one nitpicky complaint it's that I can't direct route complex options orders.
It's not impossible to make money though. When it comes down to it, the trade-off for the hours you put in and the money you make only really pays off if you're trading with someone else's money, and lots of it.
You'd get really mad at Anne Hathaway
http://www.theatlantic.com/technology/archive/2011/03/does-a...
Longer answer, this has been done for 10+ years. Bloomberg will sell you a sentiment annotated news feed for 5 figures a month if you'd like to try.
In regular trading(algorithmic or not) you really need some non-trivial insight that for some reason no one else will have.
I should probably nuance my statement that it is easy to find trading strategies by saying that it is easy to find new trading /ideas/. There are a huge number of freely available trading ideas on forums, pre-print servers (arXiv, SSRN), blogs etc. The trick is knowing how to implement them properly, accounting for any transaction costs and adjusting the parameters of the model. This is often where the stated performance falls down. It takes a lot of time to carry out this sort of research.
Long-term profitable strategies are tricky to find, due to the ever-present spectre of "alpha decay". This is where your strategy's edge is "arb'd out" - everyone else knows what you're doing and so there's no tradeable edge anymore. Hence it is necessary to have a portfolio of strategies and gradually phase out the ones that aren't doing well, and bring in new ones over time.
That being said there are a large number of trend following funds (known as Commodity Trading Advisors, or CTAs, in the industry) that all broadly do the same thing (follow "trends" in the commodity futures markets) and have great years every now and then. There are some well-known "retail" quant traders who do well by trend following, but it does require quite a bit of capital to trade in futures.
The philosophy that I do try to emphasise is to always be learning and researching new ideas. Also, as you mention, I'm pretty keen on discussing the math(s)/statistics aspect because once you have a solid math capability, it is easier to see where potential edges might exist and how to really assess whether it is a true "edge" or just a statistical anomaly.
I believe someone else in a grandchild comment below said that there are many areas that bigger quant funds won't touch because of institutional incentives. If you have $10bn assets under management (AUM), then you're not going to care about investing $100-200k, even if the returns are good, because it won't move the needle on your monthly reports.
The trick is to niche down into markets that you can spend a lot of time researching to find a distinct edge, that won't likely be touched by bigger funds. One area that is becoming interesting recently, due to the prevalence of satellite data/AI/deep learning-esque VC-backed startups, is building commodity supply/demand models. A good example is forecasting oil supply/demand by analysing large quantities of storage tank heights in global refineries [1].
Also, a small related-to-Zipline plug: I've recently started a free Python-based MIT-licensed open-source backtester [2], predominantly as a learning tool for programming and quant trading. There's about 4-5 of us working on it at the moment and it's in an early alpha stage, but we're always looking for people willing to help.
[1] - https://orbitalinsight.com/solutions/ [2] - https://github.com/mhallsmoore/qstrader/
I mean each day 100's of Phd's start with clean market
data, more data sources than you could possibly think of
and statistical back testing systems that have 1000's of
man hours put into them, trying to find a way to make money.
Is the choice really that one can be in the cohort you mention, or you can buy an index fund (or whatever the equivalent is in the market you're interested in)... and that's it?[1] Is the market so efficient that there is no middle ground where a smart and methodical person can make more money than the index plodders without being obliterated by the big players?
That seems really unlikely to me.
[1] (I don't think that's what you are saying... but you've given me a chance to try and express something I've been thinking about. Thank you for that)
You're not wrong to think that, if there's any way to consistently make money by trading, it's by exploiting inefficiencies in the market. But remember that those inefficiencies come from people. Are you confident enough in your skills, your education and your resources to be sure that you'll be one of the ones finding and exploiting inefficiencies, rather than one of the ones who's creating inefficiencies for others to find and exploit?
Yes, but then... "Finance novice beats hedge fund pros, winning $100k in Quantopian trading contest"
https://pando.com/2015/03/05/finance-novice-beats-hedge-fund...
Thousands of brilliant people are out there undiscovered and I think quantopian does a great job leveling the playing field.
If it was strongly strategy-dependent, best-information would win predictably and consistently.
Regardless of your objective - hosting a contest like this or allocating to individual institutions and strategies - you'll always face this problem. Good allocation decisions are tactical, taking both the logic of the strategy and the broader portfolio into account. Bad allocation decisions chase returns.
[1] https://en.wikipedia.org/wiki/Multiple_comparisons_problem
[2] https://en.wikipedia.org/wiki/Regression_toward_the_mean
Same rules apply to your example.
Not sure if you know, but that winning algorithm was taken offline within one month for poor performance. It was probably just overoptimised on past data to win the competition and then failed miserably out in the real market.
Previously there were other pyramid schemes, penny stocks, FX trading forums and so on. As those went out of fashion there is a new cohort of young people with great imagination who think they can beat the stock market. Some surely see a great potential there ready to exploit. They'd be silly not to.
Or to put it more obviously when everyone goes digging for goal, don't follow them, but start selling shovels. Then maybe take some profits and put some ads out in the saloons with stories about awesome finds of massive amounts of gold.
So I agree wrt/ to longer time periods. Most of those PhD's are probably trying to predict minute to minute moves, or daily moves. A lot of them would be out of a job if they lose over 90 day periods.
Not sure how much of an edge can be gained over buying and holding some reasonably diverse equity portfolio. If anyone is thinking they'll beat some HFT hedge fund they're almost certainly not going to. If they're thinking of beating a reasonably diverse/optimized portfolio over longer terms that's maybe possible IMO but the difference isn't going to be huge. To amplify the difference requires taking on more risk, for example by leveraging, options etc.
If I want to decide whether long term investing in the US vs. Turkey, Greece, Brazil, Russia, or the UK, or gold :) it's not clear machine learning can give me useful insight. The historical data has its limits. Predicting geo-political processes and things like interest rates for the long term seems like a very difficult problem. I would suspect the machine answer to this question would be something like P/E is lower so it's a good investment but who knows. So I agree with the economic/market insight comment as well.
This is spot on (tried it once :).
ML doesn't really help much at the macro level. MPT, perhaps combined with some decent economic insight, is a lot more useful.
The reason they have clean data is that big firms have teams of people dedicated to just cleaning data.
The little guy has to spend time cleaning data even before he/she starts to compete.
https://www.quantopian.com/posts/xiv-slash-vxx-pair-trade-1
Pair-trading VXX and XIV based on the StockTwits sentiments of the SPY at market open. The backtest did really well from 2011 to 2014 with 1700-1800% return in 3 years; and flat between 2014 to present-time,
I'd really love it if people can improve upon the algo and see what people when they clone the algo and come up with ways to mitigate the drawdown's and improve the performance!
How about just investing :-)
I.e. focus on periods longer than a year, which so few people/professional market participants do. And on actual businesses instead of the crazy antics of a line.
I wonder if you could use something like Zipline/Quantopian to screen huge amounts of consolidated balance sheets for markers of undervaluation. You could reject 1000s of companies and focus your “manual” vetting on the few that remain.
If you can find the dollar selling for half a dollar and you can understand why it's selling for that price (e.g. because the entire market is down), you may have identified a winner. Then all you need is a little guts and lots of patience. And a predefined set criteria that you would constantly monitor to decide if your thesis is still valid.
It's surprisingly tough to get broad, machine-readable market data for free but there are some cheaper options. Check this thread: https://www.reddit.com/r/SecurityAnalysis/comments/2ci5du/ca...
Or you could always scrape Yahoo Finance :)
In addition, nearly all finance sites provide summaries of these reports for at least the last few quarters and last few annual reports. I like morningstar.com. But finance.yahoo.com and finance.google.com both work fine.
If you want a bunch in one shot and don't have the money for Bloomberg or CapitalIQ or whatever, I suggest quandl.com, which acts as a cut-rate data aggregator for financial data.
In the latter case, investing and looking for value, it seems to me like an amateur can even give himself a little edge over the pros.
First, it's a worse fate for most professional money managers to miss out on some bull market, than to go down together with all his colleagues.
Second, amateurs working with their own money are not evaluated every quarter or even every year. They can just wait and keep looking if unsure. There are no mandates or arbitrary limitations, so the amateurs are free to look for value wherever they can find it. They can look in places that would require more patience, or that have a bit less liquidity or some more volatility (because they will typically have less money to move in/out the stock).
Of course, the pro will definitely have benefits in terms of legislation, taxes and lowering the costs of research vis. the amount of money being invested.
But in the end: does the extra information and non-GAAP stuff that the pros use, help so much over common sense and a decent understanding of accounting? (That's a genuine question, not a statement.) In fact, does succesful investing even involve outsmarting everyone else in the same way that trading does?
Primarily the artifact of sustained positive returns recently and short-term memory. There used to be a saying: When your cab driver start giving you stock tips, it is time to bail out of market. When every Tom, Dick and Harry think they can beat the market, time to take a break.
> How about just investing :-)
This is the right way to go for majority of your portfolio. Follow simple, tried and test strategies - buy Index funds/ETFs for majority of your portfolio. Bogleheads Wiki https://www.bogleheads.org/wiki/Main_Page is a good starting point.
If you really interested in individual stock/investment picking, have a very small portion of your portfolio as play money for such endeavors.
> I wonder if you could use something like Zipline/Quantopian to screen huge amounts of consolidated balance sheets for markers of undervaluation. You could reject 1000s of companies and focus your “manual” vetting on the few that remain.
I primarily use similar methodology. Automated filtering of stocks to find a few that I want to review further. It is not scalable. Majority of time is spent on developing strategy for filtering and selecting the stocks for review. I most probably manually review 15-20 stocks a year (Reading SEC filings for the company and competitors, industry news, trade articles, analyzing financial statements, etc) and invest in 3-6 stocks a year at most.
If anyone comes back to this ever, still interested in knowing!
This zipline thing is quite interesting if you're new, but if you can code, I'm not sure what the advantage is. The idea of a backtest is quite simple, and you can easily fire up something like pandas to do it for you. The equity line is simply your positions x returns, minus costs. To determine your positions, you have to make sure you aren't looking at future prices, but apart from that you are flexible in doing whatever you like.
And this was a question for me. Suppose I want to code a cross-sectional strategy. How would I do that in zipline? It seems to be the kind of thing that gives you one backtest for one time series. Perhaps I just haven't looked into it enough. When we backtest, often we want to do things across the ensemble. We also take positions in a whole universe of instruments, so the backtest needs to be a matrix, rather than just one column.
Incidentally, the example strategy will work quite well for retail traders. You can add a bunch of futures together and get a sharpe well over 1, basically what every CTA does but won't admit to. If you're wondering what all those PhDs do all day, it's adding capacity and researching minor improvements on that MA strategy. A colleague of mine worked at one of these brand names, and another friend owns one.
So, does that mean anyone can simply do this? Well, yes. But you'd have a lot of leg work to do, and you might get discouraged before you start. You need an account from someone like Interactive Brokers. You need a fair bit of money, or you'll have increment problems trading the large contracts. And you'll have to set up all the data feeds and look at it each day.
- Getting the data into a shape that you can use. Normally a total PITA. For futures, you have to either stitch the contracts yourself, or get a pre-stitched series, which you have to take time to understand. Filtering it for weird data points.
- Writing the strategy / backtesting code. The fun part.
- Connecting to a broker. Gotta read API docs, test the functions, connect it to your code in a way that makes sense, and probably in a way that makes it easy to switch brokers. Test the price feed, write error handling code.
- Daily operations code. You'll need a daily process where you can see what's going on. Automated testing of the trade report for correctness. Notifications from brokers need responses, you need to post margin as well. Some kind of SMS or Whatsapp for when something is wrong. Holiday calendar.
250k is not enough. Some of the futures contracts are quite large, and you won't be able to get the full benefit of diversification if you don't have a bunch of instruments to trade (look for a blog called Investment Idiocy, he recently talked about this). Above ~$3-5M, it isn't a problem and you can ignore it.
HFT operates on algorithms that mostly involve making money on the spread by running ahead of the brokers, buying the cheap stuff, and selling it to the broker who needs it. They don't trade on market microstructure, mostly because all of it starts to fall apart at the tick level.
250k is CERTAINLY enough to invest in futures contracts. You can do it with much less. Much much less. Futures are highly leveraged instruments. You can diversify by trading multiples of futures contracts (or e-minis depending on account size) because you're only required to post initial and maintenance margin.
You can easily blow up your account, but if you're just TRADING 5000 is enough to start selling a few contracts. If you're looking to start building something sustainable 10k is enough. But, the more the better.
Running an algorithm for multi-day trades for more than a few months does not make sense on how the markets move, as certain events like "brexit", earnings, M&A, etc... affect stock price.
If you are really interested in algorithmic trading, and you have programming experience, it's best to build your own backtesting system with intraday market data (pay for this).
This way you will know the ins and outs of a trading system.
Live trading on IB, Tradier, FXCM, Oanda and paper trading.
Local charting built in for desktop and backtesting.
Lots of tools provided for free data downloads to work with public free data libraries. lean.quantconnect.com
(I'm founder of QC :))
I realize I can't win you over but I wanted to present a fair comparison for others ;)
Also, I work on zipline
All the tutorials and examples for Zipline use daily data because there's no freely-available minutely data that we can distribute to our users.
Play with the data all you like. Don't try to trade on it if you don't really know what you're doing. (Or, just recklessly trade other people's money. It's fun.)
What you're seeing here is the "napsterization of finance." (Google it, it will lead you to the article I am almost plagiarizing).
Basically, the market at large puts together a pot of money (called "alpha", debatably) The better you are at trading, the more of that pot you get.
BUT this is not a zero sum game. It's worse.
If the markets are functioning properly, then the better you are a this, the bigger the share of the pot you get, AND the smaller the pot of money gets.
It used to be that middlemen like the NYSE stock market specialists made very large amounts of money doing what Homer Simpson automated with a drinky bird. Now, the also shops have already shrunk that pot considerably. Good news for your pension fund. Bad news for you if you try this yourself. So don't.
All the very successful day traders I know lost lots of money in the beginning before learning how to do it properly. You need to have a solid source of funds to fuel your learning, and tremendous patience.
If someone really does have a successful trading strategy, the only way it makes economic sense to publish it is if they believe they can make more money by publishing it now (i.e. selling books, pageviews, whatever) than by using it to trade. Either that, or the algorithm is being described in sufficiently general terms that you're not actually given enough information to use it effectively.
Once upon a time (1986ish) the equities and bond trading world was run by humans talking to humans and agreeing deals, the prices then fed into computer systems and the exchanges passed the prices around to make things mostly fair.
Fair of course is relative, the Eco-system was very hierarchical, with major institutions at the top, trading between each other at low fees, with brokers feeding up into them and retail shops feeding into major brokers. The customer got a raw deal, being charged heavy fees per transaction, and getting a poor "spread".
Spread was where the major institutions made their money. Human traders effectively bought very low and sold very high - both because they were human and could not easily handle algorithms in their heads and because who was going to stop them? At the top of the hierarchy traders got to see both sides of every trade - they could net trades off one against the other to make deals with little risk. And if it was not visible in a fair exchange they had even more leverage.
Spreadsheets took off around now, making it possible for one trader to plan and monitor his trades and look really good to his boss.
And then it became obvious that having a human in the spreadsheet-to-trade loop was sub optimal. A human with a spreadsheet still needed to dial a phone, make a decision, go to the toilet. A perl script could out perform him.
And at the time the algorithms were simple. If Exxon's share price dropped then pretty obviously other oil companies would drop too, but so would say car company stocks, but maybe coal miner shares would go up. And that's just in LSE - the same goes for Hong Kong and Chicago. Those correlations I could work out in a perl script. (OK, 1980, maybe some Basic :-)
And so algo trading was feasible with really tiny hardware - because the correlations in the world markets were simple, and large. And so low latency trading started. Because if I can use my ZX spectrum of my Commodore 64 to beat major traders to the punch, then all you need is a faster computer than the commodore and you beat me to the punch. And so it goes.
Fast forward twenty years and
- the hierarchy of the past is mostly still in place. Retail shops pull in the customers money, pass it upwards to brokers and they deal with traders at large banks. However the traders are much reduced, the volumes they do are orders of magnitude larger now.
- the spread has gone. Major institutions make money on tiny margins and tiny fees and just do vast vast volumes. Major FX desks will make maybe 10 USD on a billion dollars of Eurodollar trades (I think).
- the spread has gone for the algo traders. The reason PhD's are needed is because the correlations and arbitrage is all eaten up. The wins are few and far between and mostly need real world events (Brexit)
- this is generally good, there is more trade on open exchanges (good for everyone) there is smaller spreads (good for customers). The break neck automation to a good for contractors like me :-)
I'm not sure where I am going with this to be honest - but mostly it's that I am sure zip line is a good library, that the core part is written in the way a proprietary engine would look if someone took a year to rewrite it, but the core tech will not give you any edge - that edge has gone. The correlations have gone except in esoteric areas.
If you want the edge, you need to be at the top of the tree again.