FXV trade decisions

Some of you might have seen previous iterations of this project. I ditched it after I realised it needed a hell of a lot more work than I thought to be useful outside of me and my trading. What is it?

A way of selecting the best pair, best kind of trade, backed by rigorous statistical modelling. Effectively, live probabilities of price moves according to time of day, events coming up, pair, recent shocks. It answers specifically questions for FX intraday traders:

- What style of trade (trend, reversion, scalp, breakout) is most effective right now on which pairs
- What size stop over the next hour should I use (tight/medium/loose)
- What's the current condition of the market, together with markers that tell you when it's likely to change (aka regime)
- What's the likely move of the day? And the next hour?
- When should I NOT trade

All of this is done using a combination of mathematical models used across institutions worldwide - market makers, hedge funds, pension funds et al. It amazed me no-one has done this before for retail, and then when I started it I realised why. It's phenomenally complex. So complex, several times I nearly threw in the towel but eventually persisted to this, an MVP.

What it's not is a strategy. Or a buy here sell here mystical black box. Or AI anything (those who've tried, already know - AI is not yet good at structured data). And any old doughnut can wap together a bunch of labels that says BREAKOUT NOW! but that's not what this is about. It's about monitoring volatility - both realised and forecast, establishing differentials to tease out changing regimes which are invisible to the naked eye. I defy anyone to figure out the vol is 40% below normal on a chart. Everyone thinks they can, because they confuse direction with vol.

And finally, I hate dashboards - although this is one. Really hate them. Noone's ever made money from a dashboard. But it was necessary to expose the stats to ensure they're robust. And I've made the outputs - what you see on a chart - crushingly simple. Because when we're trading - me at least - I don't want to think hard about what does x/y/z indicate, I need to trade. So that's my focus and I hope it'll resonate.

I will be calling for beta testers in the coming weeks but going to use it myself first to tweak it until I know it's ready for real world use. Thus this as a journal for now. Also I realise that heavier quant stuff of coverage, qlike score and z tests is about as appealing as watching paint dry for most; basically, it's nerdy but it's necessary to get things as accurate as possible.

And for now, if you're confused as to what this really is - it all boils down to a chart and a decision matrix. On the chart, the gold box is the forecast range, the boxes are the hourly moves including what was historically forecast. Blue boxes are the future ranges - NOT where price will land. It could go up, it could go down. Imperfect but a damn sight better than finger in the air.

I trade daily so my benchmark of when this is ready for others is me being 100% sure it makes money. To do that, it needs real world testing first - ie live trades. Coming up next.

A bumpy start - because while the daily and hourlies are great, and vol regime is solid. And trend regime. The other three labels - speed, pressure and stability which are newer, made zero sense to my read of the market. Mainly disorderly all day - which it really wasn't. And so to some heavy quanting to ensure proper state separation,reworking the logic of stability and to ensure market behaviour is genuinely different between different states. Long story, short version:

GPT assessment of results
Area Verdict
State distributions Good
Behaviour separation Real
Archetype logic Encouraging
Stability logic Working
Breakout detection Surprisingly strong
Mean reversion classification Strong
Avoid classification Needs refinement later
Overfit risk Moderate but manageable

I did say it was going nowhere till i proved it myself. Already it's vastly more credible outlook. Let's see how it performs.

One more roadblock. Two things were bothering me and were obviously wrong. When during the course of a day volatility is underconsumed or overconsumed, when and how do you condition it to adjust for the impact of it. and what IS the impact of it? everything I tried was too rough, and the answer?

Surprising. as I work intraday, I'm used to everything been marked to now. It took a lot of convincing to work on hourly states for this dash, because my moves are sub 2 hours, so why so laggy? But it's how the market works - less is often more. Similarly with vol.

If Asia is quiet, NY and London will 'catch up' with higher per hour vol, still totalling the overall forecast. So the shape of the day changes. And vice versa. Wild asia, quieter London and NY. This impact is so pronounced it's almost like clockwork and if you're into options has to be the biggest cause and effect you can find of IV.

A one-off adjustment at end of Asia beats any hourly changes, hands down. Massively.

second tedious thing, I made a bad design decision - GPT guided naturally, where I combined lowvol and mediumvol. There was seemingly good reason for this but my instinct was right; it's non-sensical. The time it takes to undo this is ridiculous but now done. It improves virtually all layers by separating it out so time well spent.

these are the last of the big changes I needed, Monday apply to live....

And seven more. But some new bloody good discoveries. So to recap - think strategy + 3 more layers - regime, daily vol and hourly.

As everything works fix to fix (4pm London) I'm already half way through the day (weekends totally discounted). Which means I can get ahead looking at tomorrow am. We've still got Asia to before I start trading, so I won't look at regimes, only two things today: daily vol and ignition and exhaustion. This gives us pair selection and potential entry zones. Still rough and ready until we're in the moment.

Daily vol. So a good model tells you where price is likely to shift hard in advance. They're around 20% out on a daily median basis, so not breathtakingly accurate but compared to an ATR between 30-50% better. And these figures I'm working to are as good as any desk - you'll understand why I'm making that claim later. Also, on 5% ish of all days, a tear - they break dramatically. I've shown eurusd here as an eg. No escaping it; it's built on maths not magic.

So what's happening Monday? the model works fix to fix, ignores weekends, so we're about 8 hours through the 'day' starting Friday 4pm. Asia still to go of course before I jump in mon am.

EDIT! Ah the joys of a weekend only bug. Realised values working off different timeframes, natch. Never spotted before as it self-corrects Monday am, and don't usualyl fanny around Sunday. So revised here's the figures. We are NOT about to reverse across the board - if the previous figures were right, then yes, but as it stands no.

so big movers now - GBPNZD, CHFJPY (dangerous), EURNZD. Second tier GBPUSD.

very quiet day relatively speaking . Now to zones

One trend looks identical to another, but if you have good quality vol information, you can see one is likely spent, another is likely to continue. In plain trading terms - this is support and resistance that will break or succeed. There are exceptions to this, so this isn't hard and fast but... if you know likely vol, you can see each hour in under or over consuming. Over a given day it nets out to close to zero; one of the reasons you know your forecasts are accurate.
apply this to a trend and you can see if it's accelerating or decelerating.
find a period of 4-5 hours or more that corresponds with a trend and mark up start and finish. If it's a decelerating trend, you fade. Accelerating trend - breakout and back to base. S&R level 2 basically. I'll show an example then the future

This was AUDUSD - a peachy example by chance and not cherry picked. they won't all be this decent for sure. Because my screen is so big maybe screenshots don't cover the indicator which tells you rising or falling so FML, I've put in some arrows. So basically, decelerating down trend - this means it's spent. Those traders want out. So if it reaches their lows, you buy, if it reaches their entry point you wait to breakout, fallback and buy again. The bottom level S&R traders would all be rejoicing, the top one - they'd be shorting for sure No more.

I only ever look back up to 2 days and no more and there must be gaps of at least 4 hours before revisits; this is simply market mechanics; supply or demand needs time to replenish.

Fecking harry hindsight traders, look at my shizzle, usually right before they ask you to join their trading room. Don't worry, there is none here.
now let's look at likely candidates for Mon:
I'm not drawing effing arrows for all of them, I'll mark up the trend segments.
1) GBPNZD. the opposite - an accelerating trend - so join in when you can. We've missed this, but tells one thing, down trend as alive . DO NOT FADE! continue only.

CHFJPY. This looks more confusing but it's not really
accelerating uptrend.
From when it starts the trend drops - call this the fuckery zone. I've marked it out the big pink rectangle the main area, the top continuation zone. But it didn't continue! that's because it's late on a friday and generally trends don't (that's experience talking not hard and fast stats). It stopped at exactly 50% of expected vol, so entirely unsurprising. What it does mean, if it drops, long around that trend start but stops wide enough for the fuckery zone. and if it breaks higher, join in. Trend is looking alive and well.

GBPUSD. So you dont think is holy grail TM, here's GU.
Strong candidate for continue (acceleerating uptrend), and it did, but not from the high, but from the low of the hour that it finished (the dotted red line). This is historical so not a recco but interesting to note.. Looking back you can see there was a sudden thrust up from that end point which was a pointer that the high might not be the high, so it's nuanced.

To monday though, we finished on another accelerating uptrend, bit smaller not quite as dramatic, but we should be up from here monday.

And that'll do till tomorrow I think - you get the idea

what a shit show. Luckily, and I've never said this before I was called away for duties, because this turned out mostly awful.
The good - vol is working well, GBPNZD is on a the predicted tear today. CHFJPY same. So vol = working well.
Accelerating decelerating trends? Abysmal. Why so?
Looks like any failure and the more recent the more important invalidates the trend. On GBPNZD the indicator for reasons no idea didn't print the most recent history shown above. THis shows said history - literally a 5 hour decelerating trend meaning the downtrend is over. Most recent = most decent, clearly. In terms of entry points though it was useless (20 pips off) so I wouldn't touch it with a bargepole personally. It'd been better off using it for direction bias only then bog ordinary S&R off nearest swing points.
CHFJPY - the friday failure which I explained away, should have kept.
GBPUSD - obviously in opposition to GBPNZD but I kept it none the less - well, it did climb a little. Hardly wowsers.
Anyways from this inglorious start and not even got to entries and regimes and what not, we restart tomorrow

Its 2026 and the days of doing anything by hand are gone. I do forget, and given the rest of dash is made of stats, why I thought it was good to chart read this I don't know. I saw dozens and dozens of examples to must be true right? No.
So to concrete rules and a test across 7 pairs. Only 20 days for each. Turns out I'd visually fooled myself. ZERO impact. Not only that, but I bought into textbook definitions of this should happen because x and y. Elementary.

But. Out of every disaster... I worked out where it is useful

If the last hour trends in either direction, what's the chance of next repeating? 54%. Not much in that.
But if the last hour trends (measured by efficiency), and consumption is increasing within a certain band, what's the chance of next repeating? 62%. Next hour 61%, then 58%, then 56%. Across 8 pairs so far - it's persistent.

Proper edge. I've incorporated into existing regime labels to improve continuation ID of market. And binned of its display on the chart, noone needs to see it.

Turning to tomorrow's action, looking like a repeat with GBP, NZD, CHF leading the charge. They'll be a few actual trades using it this time rather than some shonky charts

Broadly price is moving in a sideways direction, consolidating between the resistance zone at the level around 1.3652 & the support zone which is at the level around 1.3175.
Price is moving downward after retesting the Resistance zone, in a declining channel, respecting the lower and upper boundaries of the channel multiple times as marked by Green and red arrows.

Most Recent rejection from the upper boundary of the channel as marked by yellow arrow, has led us to think of the move that can lead the price till the confluence of lower boundary of the channel & Support zone.

Emergence of a symmetric Triangle acting as a consolidation formation before breaking in the downward direction, validate our outlook!

To capture this expected decline we can plan for a short trade

High risk short entry @ 1.3402
Low risk entry @ 1.3363
Stop loss around 1.3488
For the target of 1.3200

Strict Risk management should be applied and Position should be sized according to individuals risk appetite.

For Educational Purposes only, Not an Investment Advice

That squire is 100% opposite of how I trade. I wish you well.

More false starts than a coke-fuelled track racer. If you're reading this, bear with it. It's lower priority than my day to day, and unless it adds real value I'm not deploying it to anything. the problem was a more detailed quant eval of the regimes saw their separation of state was only about 20%. Might be good enough for some, but to me felt fairly pointless, like I'd created a giant 60k lines worth of Wonka toy.
However, the dailies and hourlies were - are - clearly valuable. So I reworked around it, and came up with some genuinely impressive, persistent alpha around consumption and market mechanics.
In short, the dash gives now gives live probabilities of trend or reverse. Qualified and directional live to price. At the moment it's only backed by 20 day's historical across families of pairs but looks good enough that I'll eventually move to 2 years' pair by pair.
Plenty more to do around boxes placement, expansion/contraction but mostly there. Hell of a lot of work for this, the final output.

Wankers lingo. I'm not sure who the wankers are in this case, but here's what I do know. Everyone institutional talks to retail like they're a bunch of morons who can't tie their own shoelaces and thinks everything they're doing has never been spotted before. Similarly, you get retail shills throwing in some words to make something very simple seem like it's got legitimacy.

You never needed to know the difference before but now, if you don't want moronic answers in AI you need to use the institutional wording to get results. So a quick translation table :

Retail concept Institutional / quant equivalent
Supply & demand zone Liquidity modelling / latent order-book imbalance
Support & resistance Liquidity concentration / price-level clustering
Breakout Volatility expansion / regime transition
Fake breakout / stop hunt Liquidity sweep / stop-loss cascade
Order block Inventory transfer zone / execution footprint
Fair value gap / imbalance Thin-liquidity gap / price discontinuity
Market structure shift Structural break / state transition
Accumulation / distribution Inventory build-up or liquidation
Trend following Time-series momentum / directional drift model
Overbought / oversold Statistical deviation / z-score extremity