International markets traders banter

[Joules MM1]

Anyone think volatility is cheap as as all f right now? iVolatility has their IVX for SPX and ES running 13.~ with the 1Y low at 12....

Meanwhile:

View attachment 48610

I'm not bearish but might be time to start buying a few OTM puts and collar up momentum trades on the NYSE open, not game to exit the underlying positions but tightening what I'm expecting a little.

ops exp thursday coming for US

 

[sinner]

Sinner how many different instruments and systems do you trade?

There are seven EOD models/systems which I currently use, usually only one or two are actually running open positions at any given time depending on the volatility regime and trend. After a lot of research I am trying to reduce this down to just one or two systems all together that can handle regimes using options or futures.

Instruments for EOD trading is almost entirely equities and index futs and ETFs.

Trend following: ASX equities
Momentum: ASX equities and a different momentum system for asset class ETFs (stocks, commodities, currency, credit at the highest possible aggregate levels)

Mean reversion: Basket of international indices.
Machine learning: SPY, QQQ, DAX generally but it can be plugged into most instruments and occasionally I will do that. I might enter on the futs if I miss the close or deem the futs price to be advantageous.
Pairs trading: NYSE/NASDAQ ETFs (I admit this is mean reversion but I'm using a different model for the pairs)
Delta neutral: long straddles in a small basket of NYSE ETFs

I know it sounds like a lot of work, but all the algorithms are very fast and robust, they spit out decision tables one day ahead so basically I am just looking to execute Market on Close orders around the EOD as specified by the program.

I execute with MBTrading for everything except the ASX stuff which goes through Comsec. Size is very dynamic but on average I'm allocating about equally between short term (<5days) and anything longer than that.

As an example, I've been running only machine learning since June and started running some momentum in July.

Also I like to think I'm an intraday gun, so sometimes foolishly play the intraday FOREX (usually AUD or EUR around their respective open) and there are two components of the machine learning system (sentiment and vol) which are provide my daily 'bias' rather than signals directly. I suck at intraday mechanical FOREX but seem to be able to scrape extra pips trading discretionary. I trade breakouts and fades around the hourly highs/lows.

EDIT: I also run the asset class momentum system in my super.

 

[CanOz]

Another warning sign?

 

[CanOz]

Which way will it break? I'm thinking up(probably break down!)

 

[Punta]

basically I am just looking to execute Market on Close orders around the EOD as specified by the program.

If it's properly mechanical, why not just code a bot that does this for you, and you can log in every so often (from the pool bar) to check how it's getting on?

 

[sinner]

If it's properly mechanical, why not just code a bot that does this for you, and you can log in every so often (from the pool bar) to check how it's getting on?

My day job is in systems Punta, so I know from personal experience that there is no substitute (at least, that I can afford) for human execution.

What happens if the algo disconnects or faults at the moment it's supposed to place a trade? Or worse, exit a trade?

Basically I don't trust the computer to do that part of the job, and prefer to be fully in control there.

 

[Punta]

My day job is in systems Punta, so I know from personal experience that there is no substitute (at least, that I can afford) for human execution.

What happens if the algo disconnects or faults at the moment it's supposed to place a trade? Or worse, exit a trade?

Basically I don't trust the computer to do that part of the job, and prefer to be fully in control there.

Fair enough, it probably comes down to personal comfort. Personally, I think there's ways around these problems, such as ensuring that you always have stops lurking etc.

Actually, you should be able to make your system robust to most "conceivable" mishaps. E.g. you can have your algos only in certain "states". They do a periodic check on your portfolio, and if it doesn't match the expected state, they start exiting positions, and phoning/emailing/SMSing you.

But yeah, probably comes down to personal preference.

 

[sinner]

Fair enough, it probably comes down to personal comfort. Personally, I think there's ways around these problems, such as ensuring that you always have stops lurking etc.

A stop is just a market order. Depending on the market you're trading, and your personal experience of 'black swan' days is usually the factor on how much you trust a standing market order to get you out at the desired time, price and volume.

I would say stops are trustworthy when the market is functioning fine, but this is precisely the time that I don't need stops!

Actually, you should be able to make your system robust to most "conceivable" mishaps. E.g. you can have your algos only in certain "states". They do a periodic check on your portfolio, and if it doesn't match the expected state, they start exiting positions, and phoning/emailing/SMSing you.

But yeah, probably comes down to personal preference.

Not to sound rude, but who cares about conceivable mishaps? Those aren't the sort of problems that concern me. It's about being robust to the unexpected and unintended outcomes. My experience is that the only truly robust way of accounting for those is to size your positions properly.

Like I said, I work in systems, so this colours my view somewhat. I know exactly what true reliability is, and know that without a few million in infrastructure I could never hope to provide it. Considering human execution is free and takes about 10-30 mins per day...

 

[Trembling Hand]

A stop is just a market order. Depending on the market you're trading, and your personal experience of 'black swan' days is usually the factor on how much you trust a standing market order to get you out at the desired time, price and volume.

I would say stops are trustworthy when the market is functioning fine, but this is precisely the time that I don't need stops!

Most futs markets don't really have stop orders. IB and most brokers hold them as simulated at market orders.

Far from fail safe. And some exchanges, like Hong Kong you actually haven't even got market orders they are Limit + some amount that I cannot remember.

 

[sinner]

Most futs markets don't really have stop orders. IB and most brokers hold them as simulated at market orders.

Far from fail safe. And some exchanges, like Hong Kong you actually haven't even got market orders they are Limit + some amount that I cannot remember.

Exactly. Lock limit moves are another good example, where your stop order simply becomes another order in the queue of people trying to sell into a bidless market. What good is that? You're almost certainly never at the front of the queue.

Never happened to me, but I've imagined it, what happens if you get a margin call on a lock limit move? How can the broker exit your position if you refuse to stump up margin? It can't, best it could do is hedge in options or futs.

This is the source of the traders term "getting carted out in a bodybag"....lock limit against your margined position.

 

[Punta]

Most futs markets don't really have stop orders. IB and most brokers hold them as simulated at market orders.

Far from fail safe. And some exchanges, like Hong Kong you actually haven't even got market orders they are Limit + some amount that I cannot remember.

Sure, but if your stop doesn't work because the market is fuggered, a properly written algo will still get you out faster than a human (by re-submitting orders).

 

[Joules MM1]

Sure, but if your stop doesn't work because the market is fuggered, a properly written algo will still get you out faster than a human (by re-submitting orders).

how is an algo stop any diff to any other stop?

 

[Joules MM1]

Sure, but if your stop doesn't work because the market is fuggered, a properly written algo will still get you out faster than a human (by re-submitting orders).

if you repeat the stop placement by an algo, regardless of speed, it's no diff to moving the stop as a discretrionary and writing off more cash in the process

 

[Punta]

without a few million in infrastructure I could never hope to provide it. Considering human execution is free and takes about 10-30 mins per day...

I don't quite get this. If your algo can't communicate with the exchange, then you the human can't either. There's no loss of functionality in going to automation. What's the few mil going to get you?

To be fair though, I was only suggesting it because your systems mostly sound like EOD, which should be particularly easy to code, and easy to check at a glance that everything has been submitted correctly.

I run algos because the trading I do requires decisions/executions faster than I can think and click buttons. (They also don't get bored waiting for the right intraday patterns.) So my trading requires automation, and some risk of system failure is unavoidable I guess.

 

[sinner]

Sure, but if your stop doesn't work because the market is fuggered, a properly written algo will still get you out faster than a human (by re-submitting orders).

Look, no offense, but there is no way you're going to convince me. I work in the field of systems, I'm "academically qualified" and have been working in systems for my entire career. We are talking big systems, which run millions or in some cases billions in business.

When it comes to system failure, I've seen it all.

Here is a post from howardbandy in one of your own threads, which I found myself nodding to as I read it and found myself wanting to paste in this thread right now

Greetings --

If you are thinking about having an interface between your signal generation program and your brokers order entry facility, I recommend being very careful. In the US this is often referred to as "automated trading" or "a real-time trading interface", or something like that.

A few years ago I was a research analyst for a Commodity Trading Advisor (CTA) company and we had automated trading. We had live personnel at the computers every minute of every day that an order could be generated. We had duplicate / redundant everything -- computers, power sources, data feeds, etc. We had excellent relationships with our brokers, so that when a trade was entered due to an erroneous tick, we could make a telephone call to the broker and get out of the trade as soon as possible. We tested everything regularly, and needed it more than once.


I do recommend that trading systems be formula-based. But I recommend that the orders be placed manually after a person has verified that the trade is OK.

Thanks,
Howard

Listen to the man, he knows what he is talking about.

 

[Punta]

how is an algo stop any diff to any other stop?

Exactly!

 

[Punta]

Look, no offense, but there is no way you're going to convince me.

Not trying to mate, just interested to get your thoughts - I'm in the process of trying to "safe up" my algos at the moment, and always good to hear a bit of discussion on it ...

 

[sinner]

I don't quite get this. If your algo can't communicate with the exchange, then you the human can't either. There's no loss of functionality in going to automation. What's the few mil going to get you?

So if I'm hosting my algorithm on a virtual machine somewhere in NY and the network connection to the datacenter goes down, then what? You can't see how I as the human could manage that scenario, where the algo would fail?

Millions: Having a hosted algorithm automatically means I need a redundant hosted algorithm, redundant network links, redundant computer infrastructure, etc.

Free: Having my program setup the way it is means I can check what needs to be done in one concise page of of information and execute as long as I've got a working internet connection. The program is small and can probably even be run on an iPhone or something if necessary. Total cost of redundant infrastructure in this case is <500 for a netbook and USB 3G dongle.

To be fair though, I was only suggesting it because your systems mostly sound like EOD, which should be particularly easy to code, and easy to check at a glance that everything has been submitted correctly.

In your mind, it seems the concept of a

* Trading algorithm
* Risk management algorithm
* Order placement algorithm

Have all been conflated into a single thing. I do all the risk management and order placement myself.

 

[Punta]

So if I'm hosting my algorithm on a virtual machine somewhere in NY

I was just thinking of running your algos on the same machine that you use for human trading. That way the hardware side of things is equally reliable, at no extra cost?

I guess I thought from your original response that some of your strategies were purely rule based, and "could" be coded. Hence the question. I think it's a good discussion to have actually. Pros: you can sit on the beach while your computer trades for you. Cons: your beach excursion might be short lived because your computer empties your trading account...

 

[sinner]

I was just thinking of running your algos on the same machine that you use for human trading. That way the hardware side of things is equally reliable, at no extra cost?

So if the algo is running on my home machine and my network goes down I need to somehow get the machine connected to the internet while the algo itself remains in the correct state for all open trades (or write a chunk of code to constantly appraise the state of things and hope the code never breaks or has an edge case I've missed).

It's impossible to model all the potential state changes your system could go through in the event of an unexpected network failure or worse hardware failure. Everything needs to be redundant, including code that recognises the redundancies and kicks in at appropriate times.

I guess I thought from your original response that some of your strategies were purely rule based, and "could" be coded. Hence the question. I think it's a good discussion to have actually. Pros: you can sit on the beach while your computer trades for you. Cons: your beach excursion might be short lived because your computer empties your trading account...

All I can say is, being a student at the school of hard knocks doesn't come cheaply. Being at the beach worrying about the robustness of my algo doesn't sound too good to me.