Jack's Automated System Development Journal

0_0

When news hedging goes bad; how the last few hours of January sorta sucked.

http://i.imgur.com/r44merV.png


To be upfront: there are no excuses, the technical reasons and events surrounding the equity hit were entirely avoidable if I had taken better precautions or made a more appropriate choice given the risk of 3 high impact news events hitting the target pair within a 6 hour span.

No matter what the circumstances may be that led up to realized loss. I will still take complete ownership of the result.

I'm sour about it, since this morning I thought the monthly update was going to be awesome with a 23% gain to report. But it looks like the monthly update will have to wait til this weekend when I'll have the time to properly review what happened (blow by blow) so I can properly break down what's going to be done to avoid this in the future.

In the mean time, the basic run down was this: The algo thought it was in a heavy draw down thanks to a market shock post news release and started cutting losses accordingly. However, since a hedge position was on to flatten out risk during a news release, no actual draw down was taking place.

By the algo removing its positions, the account gained unintended directional risk at the worst possible time, since a 2nd news release was to take place shortly after. The algo then started trading again and reversed its previous position, which added to the directional risk on the account. Then the 2nd news release hit and didn't favour the positions at all..

The combined drawdown triggered the algo to, again, start cutting losses. And finally the hedge script (more of a conditional order based on time and price) figured out it was time to flatten itself too.

I didn't notice most of this even happening since I was at a night class (I'm taking a beginners course in German a few nights a week now, it's been fun) and by the time I could step in to manage things by hand there was a lot of damage done.

Then in trying to manage the exit positions left over from the EA by hand, I botched up a few executions since I was attempting to do so from my phone, in the middle of class, through a VPN app to my VPS server (because I had to manage the EA itself, not just the orders, else the EA would just reopen trades after I freed up exposure.) So frustrating to say the least. (My impact on the drawdown was very minimal, but it still didn't help things.)

What I should have done was recognized that my script to hedge on news wasn't ready to take on multiple events that overlapped each other (3 major ones that affected fundamentals within a 6 hour span,) and I should have just turned the EA off and flattened out hours before the first news release. not only would that have been more simple, but it was the better choice to make as a risk manager of my capital. That's the reality of it, I knew I was going to be preoccupied with a class, and have limited accessibility to the algo if something went wrong, but I felt overly comfortable with my EA, scripts, and setup (having just spent the last week testing them on demos) and figured trial by fire was a good idea.

All that being stated, the next step is to press pause, collect as much data on the event as possible, and start learning from what happened.

In the grand scheme of things, the drawdown wasn't major. My initial backtesting of the strategy tells me that 30-40+% drawdowns could happen easily if news or major market events weren't managed properly, so this isn't on the high side of that possibly either. And, the account is still in profit overall (even with a 1.11 PF after this mess,) the damage just killed January's profit, which is a setback, but not a death sentence.

There's more to this, not just 'should haves' and remarks from Cpt Hindsight, but the whole point is that there's no excuse: Today I failed to mange risk properly. The account should never be in a position like this without me being ready to pull the plug earlier on at the first sign of issues.

I have more details to review, which I'll be journaling here, but the plans I had to add more to the account given the good performance lately have been shelved for now.

*grumble*

Ok, so the EA has been off for the last week for a few reasons: Mainly, since I wanted to do a review and cleanup of my code / strategy after the issue with a news triggered hedge position that became unintentional risk from the week before last... and also because work has been keeping me busy lately so I haven't had a chance to commit the time I think the EA review/cleanup deserves. (My work just moved offices to a larger space and the transition has been a bit more tasking than anyone was expecting...)

That said, now that that I have a few moments to collect my thoughts and dive into this beast, I want to start chipping away at what needs to be done.

So first up: Analysis of what went wrong and what caused the drawdown event from earlier.

My initial assessment seems pretty spot on after I spent more time reviewing the play-by-play. I can't add much to it, besides maybe writing a few points to clarify.

The main breakdown happened from a disconnect between the primary EA and a hedge script (looped, like an EA) and the resulting unintended exposure that the market went against. For now, I'm moving back to manual handling of news events, or flattening out entirely before news should I not be in a position to effectively manage the hedge trade (like being stuck in a night course and trying to use a remote desktop over my phone while keeping up with the German lecture.)

Another change moving forward is committing my code to a longer testing period. This issue wouldn't have happened on my live account if I wasn't so rushed to put it into production before it was properly tested. Don't get me wrong, I did test it, but I didn't put the same QA work into it for potential issues as I did for the code I wrote back when the strategy was first formed. So a perpetual demo account running in parallel will be used for all future major updates before they hit production.. and I'll be looking for at least a month of forward testing through major market events as a good indicator that it's working well.

I'll also be leaning toward straight flattening out before news instead of hedging out net exposure. The main reason is there's too many unknowns that happens post news to expect conditions to be favorable to my edge even if the initial shock of the news has passed. For instance, sure I can stop a 50-100 pip snap in a currency pair from harming my position, but if the news is drastic enough to reshape market sentiment and the pair then trends hard in a new direction, my EA's previous trade has lost its edge and clearly I'm hoping on a more random outcome to let me get out of the trade at a decent price (profit or not.)

This is actually something I should have realized before. In my manual trading, both in FX and at work with equities, I don't entertain a directional bias til after I see how price reacts to news. So why should my algo and automated strategy carry a bias through news at all? So that lesson has been relearned.

Someone PM'd me recently and asked if the loss was because the target pair had a strong trend with little pullback over the last few days. I suppose they assumed my strategy just averages in against the trend or something.. which it doesn't... and most likely they were just trying to figure out what I'm doing so they could replicate it. So while I'm not about to spill the beans there, I will cite some interesting facts of the drawdown that happened to dispel this thought in anyone else reading: the core strategy itself would have actually come out profitable thanks to the strong trend, not despite it, and the majority of the losses were related to the hedge script mucking things up and causing unintended risk.

For instance:

http://i.imgur.com/DXNiVeR.png

Here we see the hedge trades singled out from the group.

And below we see the main strategy's impact on % return:

http://i.imgur.com/UaplNd1.png

Actually, related to the above (why there's so many separately tracked lines

One thing that has been dogging me lately is being unable to backtest all components of my EA at once, since I've broken functions up between different EAs on different charts and MT4's strategy tester only runs one EA at a time and can't merge results from multiple tests into one report. Right now I'm using 1 EA per order handling method (Dash, Runner, and now Nibbles,) 1 EA per direction (long, short) and finally 1 EA/script to manage news hedging... so that's 4-6 active EAs at any given time all effectively working together to be a single strategy. At first, I thought I was being smart by doing this so I could remove / add and adjust components on the fly without messing with the rest of the code, even break up tasks to other boxes should they start to be too demanding for one computer (not that my EA was very demanding, just thinking of my future plans and new strategies I want to try out that might be CPU intensive,) ...but now I see that it's holding me back greatly, since I can't effectively backtest at this point, and forward testing takes too long so I feel rushed into production for changes I want to see happen sooner.


A quick recap on what's going to happen moving forward:

 

1. EA will go back on sometime this week, but I'll be taking over the news hedging function by hand like I was in previous months.
2. The next big change will be moving all separate scripts/EAs into one monolithic EA since it's all the same strategy, but this will take some time since I'll have to redo how the EA tracks tickets and calculates it's positions. This could be a month or two out since I'm slow at programming (amateur after all.)
3. I will use this transition to a monolithic structure to build out an EA template for order management, which I hopefully can reuse when I add in entirely new strategies in the future.
4. I will strictly run all major structural changes to the EA in demo for a month (at least) before committing them to production.

PS. If it seems like I'm putting way too much effort into an issue over a loss not greater than $100, it's because I'm treating this as any serious trader should, it's a learning experience.. the money is gone to waste if I don't walk away form this having learned what doesn't work. Plus, I'm about to describe how this account will be growing rapidly in the coming months, and I imagine any loss (small or not) as it would be relative to my ideal account size.

This is akin to how I started off trading equities at work and trading FX at home years ago. Small but live, prove risk is worth taking through executed trades and performance (not theory,) and then sale up. If something doesn't work out, it doesn't hurt much at all, but if it works then raise risk and *happydance*.

On Growing the Account:


My goal is to grow the account to be many times the size it is now (which is really nothing as it stands.)

My plan is to treat the account like another 'trade' in my overall manual portfolio. I will add to winners if warranted, and reduce or cut losers.

Now, the problem is, how do I grow this effectively in a way that encourages me to keep up development and refinement of my code/EA/strategies? (After all, as fun as this project is, it does take up a lot of time and returns very little for it.)

I don't want to just toss $10k at this in case something major goes wrong, that would be misplaced risk and that capital would better be allocated to my manual accounts.

But I can't just expect the account to grow itself into $10k since that would be grossly unrealistic (or take a few decades.)

So I think I found the right balance:

I will be matching all profits made via deposit on a weekly basis. Each week, the $ value returned will be matched with a deposit (most likely a transfer from my manual account,) and if there's no profit then no deposit will take place.

I will also operate on a high watermark basis. If there is a loss from one week, I will only count new profit past the last high watermark.

This might not be a lot at fist, but it will help things grow a little faster, and as the account size grows, so will the potential 'top-ups' going in.

So..: as of last week, I put in a deposit to match the existing profit on the account, which would be $101.18

That means, the total net profit on the account is still $101.18 since inception (sour it's not near $190 like from before the spill I had) and I've added additional equity to the account of $101.18.

The target balance range for the account is between $5-10k to start. At which point I will reevaluate how I plan on growing things from there (it might not be as convenient to match profit if it's a few grand in a given week, which might happen at that level.)

If you guys agree with this idea, or disagree, or think there's a better way, please speak up.


Also, just to back up my words, you might have noticed nothing has changed on the MyFXBook tracking page .. I have no plans on trying to hide anything (including failures) and you'll see the equity has increased thanks to the deposit I talked about in this post.

EDIT: I just noticed that MyFXBook is starting an auto-trade (trade copy) service. I have no plan, nor do I foresee myself ever having plans, to offer such a service. I want to make this clear that my links to MyFXBook are for 3rd party performance trading and analysis tools only and my intentions are entirely non-commercial. Maybe I'll ask ForexFactory to implement the features I need on TradeExplorer so I can use that instead of MyFXBook... if MyFXBook is going this route that is.

Despite the recent drawdown the system remains attractive... 27% return in 6 months is awesome. At this pace, one can expect about 50% in annual return. Though some investors might find the 30% drawdown a little too aggressive. Overall an 'interesting' feat by a robot.

http://www.reactiongifs.com/wp-conte...3/03/right.gif

Thanks for the kind words.

I'm trying not to "expect" anything in terms of return. All I've done is set what I believe to be acceptable risk on the strategy's trades and let the market do the rest. The worst thing one can do is 'expect' a given return, instead, just embrace what opportunities the market gives you and don't force your expectations on it.

And yeah, the drawdown stats are hurting bad. It's really just bad timing, there was a slip up that wasn't related to the strategy itself (documented this a few posts earlier, the method I used to hedge out of exposure during news releases went haywire thanks to my noobish coding mistakes,) combined with a drawdown following that was caused by the strategy itself (which on it's own would be well within what I expect to see.) So combined we see the high drawdown % number, sadly.

I've made some changes recently and added a new strategy that runs on EUR/USD which I'll be detailing in a new post.

LOL, I take it you agree.

Thanks for your comments Jack... that is very true, what you had to say about the expectation bit.

I was reading your other thread on professional trading and found it seriously enriching. From your perspective, what is the 'maximum daily drawdown' that might be tolerated for a manual system at a professional level (for leverage less than 5 to 1)? And would that figure be different for an automated system?

Now that Jack is labeled as Commercial Member, can he still reply or post here on his own thread? I believe he would be stuck on Commercial Forum and cannot post anywhere else. And he cannot receive private message now.

Hello Reader,

The following is Jack's response to my query above:

"Ok, so I actually had a response drafted up for your question which I couldn't post before:

It entirely depends on the structure of the firm you're working for... that's the trouble with your question, what's acceptable for one firm could be something to get fired over at another. Each is different.

The largest factor is the investment goals of the firm, and who's providing the capital.

Few examples:

A bank running a flow trading operation (trading on behalf of clients) will have different objectives on a client by client basis, and a trader's risk is judged as it impacts each account separately. One client account might be restricted to a few % a year variance in performance, others might be more aggressive.

A bank / firm trading their own funds: Usually a lot more strict, a lot of time/effort goes into opportunity cost when it comes to where capital is locked up. Recently this has left most banks pulling funds from traders and putting them into other areas of the business (not to mention all the new regulations post-credit crisis making it hard for banks to trade their own capital.

Pension funds / Insurance Funds: Even stricter here. These guys are so worried about meeting future capital outlays that if you better have a damn good reason why something was worth the risk if you put a dent in their fund's capital. Most have put more effort into direct investment lately, since the markets have been hard on 'buy and hold' style long term investing lately... The Teachers Pension here in Ontario partly owns a hockey team, and bid on owning one of our tallest skyscrapers in Toronto.. these investments made more sense from a cash flow perspective than trading in the markets or trading bonds.. go figure.

A trader at a hedge fund / active mutual fund: Again, depends on the fund specifically, but usually their risk will have to follow the fund's prospectus. Generally speaking, if losing 20% in a year would be an outright disaster, then losing 1-3% in a day would be pretty bad.

A trader at a private capital firm: Each is different (I know, I've said this a few times now,) and I can't speak for all of them but for my own work I'm not allocated a % of a capital pool, I'm given a $ value in buying power and a risk limit on a per day basis I can lose at most. If I come close to that risk limit, it's my job to manage my positions so they are closed up without breaching said risk limit. My bosses don't care much if I hit my risk limit for the day, but they care a lot if I go over it even by a few bucks. Our risk limits are set so they can be recovered from within 1-3 average days of performance. Each daily limit is set per trader based on performance; the better we do, the more rope they give us to hang ourselves later.

I could keep going on, but with each type of institution / firm, the answer is always going to be the same: 'it depends on them'.

General rule of thumb, no fund/company I know of has lasted 10+ years allowing a single trader to lose a few full % points of their working capital... they treat risk very seriously, and are still alive after 10+ years of operations to prove why.

Like the old expression goes: there are old traders, and there are bold traders, but you'll never meet an old bold trader.

Hope that helps."

I have seen the result in live account of this EA. The drawdown of this EA is 34.21% which should be less than 10% I think. The is gain is 34%, which is not so large but positive. The short and long won is over 70% thats good. Last month the result was very bad. Also the result is only for six months which is too short and looking further for long term result.

Thanks for an interesting read. You might want to consider using one of the many news indicators on FF. They each read the FF news calendar and plot or otherwise use the info. It should not be overly difficult to incorporate one of these into your EA to avoid news periods automatically...just to remove some more of your manual intervention If interested you might start with Hanovers thread here. It's out of date but will lead you to others that use the current FF calendar.