the SNB floor removal: january 15, 2015 — the most violent FX move in G10 history.
background: in september 2011, the SNB set a floor on EURCHF at 1.2000. they committed to buying unlimited euros to prevent CHF from strengthening beyond this level. the stated reason: CHF was dangerously strong, threatening swiss export competitiveness and risking deflation.
the floor held for 3 years and 4 months. markets priced it as essentially permanent — brokers reduced margins on EURCHF, traders built carry strategies around it, and the pair barely moved.
january 15, 2015, 09:30 CET: the SNB announced, without warning, that it was abandoning the floor. EURCHF dropped from 1.2000 to a low of approximately 0.8500 within minutes — a 30% move in one of the most liquid G10 currency crosses.
the impact: several retail FX brokers became insolvent because client losses exceeded account equity — the broker absorbed the difference. many professional traders lost multiples of their intended maximum loss.
the permanent lesson: no peg, floor, or ceiling in FX is permanent. when central banks remove them, they remove them without warning. sizing for tail risk in any pegged or managed pair must account for this, regardless of how long the structure has held.
background: in september 2011, the SNB set a floor on EURCHF at 1.2000. they committed to buying unlimited euros to prevent CHF from strengthening beyond this level. the stated reason: CHF was dangerously strong, threatening swiss export competitiveness and risking deflation.
the floor held for 3 years and 4 months. markets priced it as essentially permanent — brokers reduced margins on EURCHF, traders built carry strategies around it, and the pair barely moved.
january 15, 2015, 09:30 CET: the SNB announced, without warning, that it was abandoning the floor. EURCHF dropped from 1.2000 to a low of approximately 0.8500 within minutes — a 30% move in one of the most liquid G10 currency crosses.
the impact: several retail FX brokers became insolvent because client losses exceeded account equity — the broker absorbed the difference. many professional traders lost multiples of their intended maximum loss.
the permanent lesson: no peg, floor, or ceiling in FX is permanent. when central banks remove them, they remove them without warning. sizing for tail risk in any pegged or managed pair must account for this, regardless of how long the structure has held.
the difference between being wrong and being early.
these two things feel identical when you're in the position. they are not the same.
being wrong: the thesis was incorrect. the driver you identified did not move price because it was not the primary driver, or it was already fully priced, or a counter-driver was stronger.
being early: the thesis is correct but the catalyst has not yet arrived. price hasn't moved yet because the information hasn't fully disseminated or the positioning adjustment hasn't happened.
how to tell them apart — this is genuinely difficult in real time. but:
— if new information has emerged that contradicts your thesis, you are probably wrong.
— if the thesis is intact but timing was off, you may be early.
— if the market is moving against you in the absence of new information, consider that you may have identified the right theme but at the wrong magnitude.
the response to being early: hold, but size appropriately so that being early for longer than expected doesn't become a capital event. the response to being wrong: exit, review, and update your model.
these two things feel identical when you're in the position. they are not the same.
being wrong: the thesis was incorrect. the driver you identified did not move price because it was not the primary driver, or it was already fully priced, or a counter-driver was stronger.
being early: the thesis is correct but the catalyst has not yet arrived. price hasn't moved yet because the information hasn't fully disseminated or the positioning adjustment hasn't happened.
how to tell them apart — this is genuinely difficult in real time. but:
— if new information has emerged that contradicts your thesis, you are probably wrong.
— if the thesis is intact but timing was off, you may be early.
— if the market is moving against you in the absence of new information, consider that you may have identified the right theme but at the wrong magnitude.
the response to being early: hold, but size appropriately so that being early for longer than expected doesn't become a capital event. the response to being wrong: exit, review, and update your model.
Forwarded from EQUILON | John Zhan
The ECB has brought together central bank governors in Sintra — yet there’s hardly any mention of it on the news channels
The ECB Forum in Sintra (Portugal) is a closed-door annual event, taking place this year from 29 June to 1 July.
This year’s theme is ‘innovation, growth and stability’. I’ll be highlighting the key points as they emerge and sharing them here.
What I’m specifically keeping an eye on:
Today (Mon):
✔️ Opening ceremony and dinner, with an opening address by Lagarde. This is largely a ceremonial event and is unlikely to have much impact on the markets.
Tomorrow (Tue) – AI day:
✔️ 11:40, ‘AI and Financial Stability’ panel: Tobias Adrian (IMF), Sarah Briden (Bank of England) + academics and Apollo’s chief economist.
✔️ 14:30, a separate discussion on AI — Aaron Chatterjee, Chief Economist at OpenAI, and Philip Lane from the ECB.
This is what I’ll be listening to first and foremost: I’m curious to hear what someone from OpenAI has to say about AI in the context of the economy and regulation.
Wednesday:
✔️9.45, a session on tokenisation. Hyun-Seong Shin (Bank of Korea) will be presenting on the Hangang unified registry project. Tokenisation is currently one of the key issues for the entire financial system.
✔️ A context that cannot be ignored: Binance recently withdrew its application for a MiCA licence in Greece. From 1 July, it will be impossible to operate in the EU without a licence — the exchange has already halted the registration of new users in Europe and is scaling back some of its services (whilst stating that clients will retain access to their funds). It plans to apply for a licence in France next. Against this backdrop, the issue of crypto regulation takes on particular urgency.
✔️ 14:00, political panel and forum closing session: Bailey (Bank of England), Lagarde (ECB), Macklem (Bank of Canada) and Warsh (US Federal Reserve).
I’m keeping a close eye on Warsh in particular — the new head of the Fed; his speech here is of particular interest.
I’ll keep you updated on any important news over the next few days.
The ECB Forum in Sintra (Portugal) is a closed-door annual event, taking place this year from 29 June to 1 July.
This year’s theme is ‘innovation, growth and stability’. I’ll be highlighting the key points as they emerge and sharing them here.
What I’m specifically keeping an eye on:
Today (Mon):
✔️ Opening ceremony and dinner, with an opening address by Lagarde. This is largely a ceremonial event and is unlikely to have much impact on the markets.
Tomorrow (Tue) – AI day:
✔️ 11:40, ‘AI and Financial Stability’ panel: Tobias Adrian (IMF), Sarah Briden (Bank of England) + academics and Apollo’s chief economist.
✔️ 14:30, a separate discussion on AI — Aaron Chatterjee, Chief Economist at OpenAI, and Philip Lane from the ECB.
This is what I’ll be listening to first and foremost: I’m curious to hear what someone from OpenAI has to say about AI in the context of the economy and regulation.
Wednesday:
✔️9.45, a session on tokenisation. Hyun-Seong Shin (Bank of Korea) will be presenting on the Hangang unified registry project. Tokenisation is currently one of the key issues for the entire financial system.
✔️ A context that cannot be ignored: Binance recently withdrew its application for a MiCA licence in Greece. From 1 July, it will be impossible to operate in the EU without a licence — the exchange has already halted the registration of new users in Europe and is scaling back some of its services (whilst stating that clients will retain access to their funds). It plans to apply for a licence in France next. Against this backdrop, the issue of crypto regulation takes on particular urgency.
✔️ 14:00, political panel and forum closing session: Bailey (Bank of England), Lagarde (ECB), Macklem (Bank of Canada) and Warsh (US Federal Reserve).
I’m keeping a close eye on Warsh in particular — the new head of the Fed; his speech here is of particular interest.
I’ll keep you updated on any important news over the next few days.
what "the trend is your friend" actually means — and when it stops being true.
the expression is correct in a specific context: in a trending market, trading in the direction of the established trend has a higher base rate of success than trading against it. this is a statistical statement about base rates, not a guarantee.
when it is true:
— the trend has been in place long enough to reflect a genuine shift in the underlying driver (rate differential, growth differential, capital flow)
— the trend is visible on at least two timeframes
— momentum confirms: price makes higher highs and higher lows (or lower lows/lower highs) consistently
when it stops being true:
— the trend has been in place long enough that the original driver is already fully priced
— positioning has become extreme (everyone is already in the trade)
— a reversal catalyst is approaching (central bank meeting that may change the rate path)
— volatility compresses and the pair stops making new highs despite attempts
trend-following works over large samples. the difficulty is that individual trends end at unpredictable times, and the exit signal on a trend is rarely as clear as the entry signal was at the start.
the expression is correct in a specific context: in a trending market, trading in the direction of the established trend has a higher base rate of success than trading against it. this is a statistical statement about base rates, not a guarantee.
when it is true:
— the trend has been in place long enough to reflect a genuine shift in the underlying driver (rate differential, growth differential, capital flow)
— the trend is visible on at least two timeframes
— momentum confirms: price makes higher highs and higher lows (or lower lows/lower highs) consistently
when it stops being true:
— the trend has been in place long enough that the original driver is already fully priced
— positioning has become extreme (everyone is already in the trade)
— a reversal catalyst is approaching (central bank meeting that may change the rate path)
— volatility compresses and the pair stops making new highs despite attempts
trend-following works over large samples. the difficulty is that individual trends end at unpredictable times, and the exit signal on a trend is rarely as clear as the entry signal was at the start.
the only question that matters after a trade closes.
not: did I make money?
not: was I right about direction?
not: should I have held longer?
the question: did I execute the process correctly?
if the answer is yes, and you lost money, that is information. your process, correctly applied, produced a loss. this happens. it is expected. it does not mean the process is wrong.
if the answer is no, and you made money, that is also information — and it is not good news. a profitable trade taken with a flawed process does not validate the process. it is a random outcome that may not repeat.
the traders who improve are those who evaluate their process rigorously and are willing to accept profitable trades as failures if the execution was undisciplined, and losing trades as successes if the execution was correct.
this is psychologically difficult because the market gives you P&L feedback, not process feedback. building the habit of evaluating process independently of outcome is the work that separates traders who learn from those who just accumulate experience without improving.
not: did I make money?
not: was I right about direction?
not: should I have held longer?
the question: did I execute the process correctly?
if the answer is yes, and you lost money, that is information. your process, correctly applied, produced a loss. this happens. it is expected. it does not mean the process is wrong.
if the answer is no, and you made money, that is also information — and it is not good news. a profitable trade taken with a flawed process does not validate the process. it is a random outcome that may not repeat.
the traders who improve are those who evaluate their process rigorously and are willing to accept profitable trades as failures if the execution was undisciplined, and losing trades as successes if the execution was correct.
this is psychologically difficult because the market gives you P&L feedback, not process feedback. building the habit of evaluating process independently of outcome is the work that separates traders who learn from those who just accumulate experience without improving.
🤖 Today at the forum was AI day
Out of the whole AI session, two speakers actually stood out — Aaron Chatterji and Torsten Slok. They prepared independently, yet their points ended up complementing each other.
Chatterji — on the labour market:
• AI won't cause mass unemployment and won't displace workers — but it will reshape the labour market.
• The parallel with the internet and electricity: PCs caused a brief dip in the 1990s, then a hiring boom followed. AI will play out the same way.
• Employment data does not support mass AI-driven layoffs. The wave of layoffs late last year had a different cause (more on that below).
• The forecasts he leans on: Goldman Sachs — +7% to global GDP over 10 years; McKinsey — up to +3.4 percentage points to annual productivity growth.
• For now, AI only replaces narrowly scripted, rules-based tasks. The job for businesses is not to fear it, but to build it into their processes.
Slok — and here's what should actually worry us:
He agrees that AI's direct impact on the labour market is limited. But the real risk isn't jobs — it's debt. And it's mainly people who actually trade the markets who pick up on this; most others miss it entirely.
• AI companies (OpenAI, Oracle, Meta, Google and others) are issuing bonds in amounts larger than they can realistically cover.
• The hyperscalers (Oracle, Meta, Google, Amazon) have already issued ~US$250 billion in bonds so far in 2026 — roughly 60% of their combined market cap.
• Almost half of all new corporate bond issuance is now tied to AI. The money goes into infrastructure — data centres, power, cooling — but AI revenue can't yet cover those obligations.
• If spreads widen → volatility in the debt market → default risk → a credit crunch → and the "AI bubble" everyone talks about could burst.
• From there → a macro shock and recession. And that's when unemployment truly surges.
Bottom line: the thing to fear isn't robots — it's the debt overhang building up under the AI boom. The main risk to the market is bonds, not automation.
I'll save a couple of the most interesting details for next time 👀
Out of the whole AI session, two speakers actually stood out — Aaron Chatterji and Torsten Slok. They prepared independently, yet their points ended up complementing each other.
Chatterji — on the labour market:
• AI won't cause mass unemployment and won't displace workers — but it will reshape the labour market.
• The parallel with the internet and electricity: PCs caused a brief dip in the 1990s, then a hiring boom followed. AI will play out the same way.
• Employment data does not support mass AI-driven layoffs. The wave of layoffs late last year had a different cause (more on that below).
• The forecasts he leans on: Goldman Sachs — +7% to global GDP over 10 years; McKinsey — up to +3.4 percentage points to annual productivity growth.
• For now, AI only replaces narrowly scripted, rules-based tasks. The job for businesses is not to fear it, but to build it into their processes.
Slok — and here's what should actually worry us:
He agrees that AI's direct impact on the labour market is limited. But the real risk isn't jobs — it's debt. And it's mainly people who actually trade the markets who pick up on this; most others miss it entirely.
• AI companies (OpenAI, Oracle, Meta, Google and others) are issuing bonds in amounts larger than they can realistically cover.
• The hyperscalers (Oracle, Meta, Google, Amazon) have already issued ~US$250 billion in bonds so far in 2026 — roughly 60% of their combined market cap.
• Almost half of all new corporate bond issuance is now tied to AI. The money goes into infrastructure — data centres, power, cooling — but AI revenue can't yet cover those obligations.
• If spreads widen → volatility in the debt market → default risk → a credit crunch → and the "AI bubble" everyone talks about could burst.
• From there → a macro shock and recession. And that's when unemployment truly surges.
Bottom line: the thing to fear isn't robots — it's the debt overhang building up under the AI boom. The main risk to the market is bonds, not automation.
I'll save a couple of the most interesting details for next time 👀
sizing to volatility: why a fixed pip stop is the wrong approach.
traders who set a fixed 30-pip stop on every trade are implicitly taking different dollar risk on every trade, because pairs have different volatility profiles and current volatility changes over time.
the better framework: size your position so that your dollar risk is consistent, and let the pip stop vary based on the current ATR.
example:
— you want to risk $200 on every trade
— EURUSD daily ATR: 80 pips → stop at 1× ATR = 80 pips → position size: $200 / (80 × $10) = 0.25 lots
— GBPUSD daily ATR: 120 pips → stop at 1× ATR = 120 pips → position size: $200 / (120 × $10) = 0.167 lots
the result: consistent dollar risk per trade regardless of which pair you trade or how volatile the market is.
the secondary benefit: when volatility rises (ATR increases), your position size automatically decreases. you naturally take smaller positions in more volatile environments — which is the correct behavior — without needing to consciously decide to reduce size.
traders who set a fixed 30-pip stop on every trade are implicitly taking different dollar risk on every trade, because pairs have different volatility profiles and current volatility changes over time.
the better framework: size your position so that your dollar risk is consistent, and let the pip stop vary based on the current ATR.
example:
— you want to risk $200 on every trade
— EURUSD daily ATR: 80 pips → stop at 1× ATR = 80 pips → position size: $200 / (80 × $10) = 0.25 lots
— GBPUSD daily ATR: 120 pips → stop at 1× ATR = 120 pips → position size: $200 / (120 × $10) = 0.167 lots
the result: consistent dollar risk per trade regardless of which pair you trade or how volatile the market is.
the secondary benefit: when volatility rises (ATR increases), your position size automatically decreases. you naturally take smaller positions in more volatile environments — which is the correct behavior — without needing to consciously decide to reduce size.
the hardest moment in any trade.
it is not the entry. entries are relatively simple — you see a setup, you take it or you don't.
the hardest moment is after a strong initial move in your favor, when price pulls back toward your entry.
the behavioral response: the gain is at risk of disappearing. the fear of losing the unrealized profit can be stronger than the original fear of losing capital. many traders close here — taking a small profit rather than the intended target — and then watch the trade continue in their direction.
why the pullback is not the signal to exit:
— pullbacks within a trend are normal. a move from entry to 50% of target, followed by a 30% retracement, is normal price action in most trends.
— your exit criteria should be defined before entry. a pullback to entry is only a problem if your exit criteria include "price retreats to entry."
the solution: define, before the trade, the specific condition under which you will exit. a close below a specific level. a specific time horizon. not "when I'm scared."
process before entry eliminates most decisions during the trade.
it is not the entry. entries are relatively simple — you see a setup, you take it or you don't.
the hardest moment is after a strong initial move in your favor, when price pulls back toward your entry.
the behavioral response: the gain is at risk of disappearing. the fear of losing the unrealized profit can be stronger than the original fear of losing capital. many traders close here — taking a small profit rather than the intended target — and then watch the trade continue in their direction.
why the pullback is not the signal to exit:
— pullbacks within a trend are normal. a move from entry to 50% of target, followed by a 30% retracement, is normal price action in most trends.
— your exit criteria should be defined before entry. a pullback to entry is only a problem if your exit criteria include "price retreats to entry."
the solution: define, before the trade, the specific condition under which you will exit. a close below a specific level. a specific time horizon. not "when I'm scared."
process before entry eliminates most decisions during the trade.
what experienced traders actually do differently: ten observations.
one: they have a specific reason for every trade. not a feeling — a specific, articulable thesis.
two: they know their stop before they know their target.
three: they check their position size against their risk budget before entry, not after.
four: they don't check their positions every ten minutes. they set alerts and check at defined times.
five: they distinguish between their thesis being wrong and the timing being off.
six: they take days off. not because they can't trade, but because they recognize when conditions aren't right for their approach.
seven: they review their trades monthly with the same seriousness they prepared them.
eight: they don't add to losing positions without a specific revised thesis. averaging down as a default is not a strategy.
nine: they are skeptical of their winning periods. a run of wins can mask a strategy that is working for luck-related reasons in a specific environment.
ten: they are not impressed by other people's wins. they know that a single trade result, or even a month of results, means almost nothing about underlying edge.
one: they have a specific reason for every trade. not a feeling — a specific, articulable thesis.
two: they know their stop before they know their target.
three: they check their position size against their risk budget before entry, not after.
four: they don't check their positions every ten minutes. they set alerts and check at defined times.
five: they distinguish between their thesis being wrong and the timing being off.
six: they take days off. not because they can't trade, but because they recognize when conditions aren't right for their approach.
seven: they review their trades monthly with the same seriousness they prepared them.
eight: they don't add to losing positions without a specific revised thesis. averaging down as a default is not a strategy.
nine: they are skeptical of their winning periods. a run of wins can mask a strategy that is working for luck-related reasons in a specific environment.
ten: they are not impressed by other people's wins. they know that a single trade result, or even a month of results, means almost nothing about underlying edge.
when the trade is right but the market goes wrong.
there are trades where your thesis is fundamentally correct — the economic relationship you identified is real — but the position loses money anyway.
possible reasons:
one: timing. the catalyst you expected is delayed by 6 weeks. during that time, the position bleeds. if your stop is properly sized, you exit and the market eventually confirms your thesis. someone else benefits.
two: the dominant driver switched. your analysis of the rate differential was correct, but a risk-off event temporarily overrode the rate story. when the risk event resolves, the rate differential reasserts.
three: the trade was correct but overcrowded. everyone had the same thesis. the initial move in your direction caused a crowded-positioning squeeze against the consensus before ultimately resolving in the direction of the thesis.
the correct response in all three cases: accept the loss, log it with the context, and update your model about timing and catalysts. do not tell yourself "the market was wrong." the market is the input. your job is to assess it correctly. if you didn't, find out why.
there are trades where your thesis is fundamentally correct — the economic relationship you identified is real — but the position loses money anyway.
possible reasons:
one: timing. the catalyst you expected is delayed by 6 weeks. during that time, the position bleeds. if your stop is properly sized, you exit and the market eventually confirms your thesis. someone else benefits.
two: the dominant driver switched. your analysis of the rate differential was correct, but a risk-off event temporarily overrode the rate story. when the risk event resolves, the rate differential reasserts.
three: the trade was correct but overcrowded. everyone had the same thesis. the initial move in your direction caused a crowded-positioning squeeze against the consensus before ultimately resolving in the direction of the thesis.
the correct response in all three cases: accept the loss, log it with the context, and update your model about timing and catalysts. do not tell yourself "the market was wrong." the market is the input. your job is to assess it correctly. if you didn't, find out why.
the final test of any trading system: forward performance.
a strategy that works in back-testing is necessary but not sufficient. a strategy that works in the last three months of live trading is also not sufficient. the real test is forward performance over a full market cycle.
a full market cycle includes: a trending period, a ranging period, a high-volatility period, a low-volatility period, and at least one unexpected event (flash crash, policy surprise, data shock).
why back-testing overstates performance:
— you selected the strategy because it worked on this data. this is data-mining bias.
— you know where the major events were, so you may have (consciously or not) avoided the periods that would have been hardest.
— transaction costs in live trading are worse than assumed.
the only resolution: forward testing with real money (at small size) over a sufficiently long period. 100 trades is a minimum sample. 200+ gives more statistical confidence.
the traders who believe their strategy before 200 live trades are almost always working from insufficient evidence. the traders who have 500+ live trades documented with a clear process have actual information about their edge.
a strategy that works in back-testing is necessary but not sufficient. a strategy that works in the last three months of live trading is also not sufficient. the real test is forward performance over a full market cycle.
a full market cycle includes: a trending period, a ranging period, a high-volatility period, a low-volatility period, and at least one unexpected event (flash crash, policy surprise, data shock).
why back-testing overstates performance:
— you selected the strategy because it worked on this data. this is data-mining bias.
— you know where the major events were, so you may have (consciously or not) avoided the periods that would have been hardest.
— transaction costs in live trading are worse than assumed.
the only resolution: forward testing with real money (at small size) over a sufficiently long period. 100 trades is a minimum sample. 200+ gives more statistical confidence.
the traders who believe their strategy before 200 live trades are almost always working from insufficient evidence. the traders who have 500+ live trades documented with a clear process have actual information about their edge.