Forem: Kenny Chan

When AI Starts to Understand Market Narratives: How the AWS Financial Services Symposium Is Rethinking Event‑Driven Trading

Kenny Chan — Fri, 27 Mar 2026 02:03:43 +0000

当AI开始读懂市场叙事：AWS Financial Services Symposium重构事件驱动交易的底层逻辑

Thanks to Jacky Wu for the invitation to the AWS Financial Services Symposium. David Sung and Michelle Hong shared industry insights on AI in the AWS Financial Services. Danny Chan supports the AWS FSI Acceleration Professional Services Group.

Key Takeaways

① Turning unstructured information into signals.

Unstructured data is converted into embeddings and analyzed using vector similarity matching. This enables semantic-level AI tagging that systematically links news, macro events, and cross-asset price movements into a signal layer. As a result, human traders can focus on the highest-value work in trading alpha: (1) building hypotheses, (2) designing trade structures, and (3) managing execution-level strategy.

将非结构化信息通过 embedding 向量化，利用 vector similarity matching 进行语义级 AI Tagging，把新闻、宏观事件与跨资产价格行为系统性地映射为信号层。由此，人类交易员得以专注於 alpha ①假设构建、②交易架构设计与③执行层博弈等最高价值的交易实务。

② Fixing long-standing problems in discretionary macro trading.

Macro discretionary trading has long faced structural issues: (1) strong path dependence, (2) limited data samples, and (3) difficulty in standardized backtesting. From a regime perspective, the macro environment of 2026 looks closer to a 1999-style regime, shaped by three overlapping themes: (1) high-valuation technology narratives, (2) rising geopolitical uncertainty, and (3) repeated testing of real-economy fundamentals. In this framework, AI tagging continuously labels (1) narrative spread, (2) risk-factor exposure, and (3) asset co-movement, providing a more stable signal layer for macro hedging alpha.

宏观主观交易长期面临①路径依赖强、②样本稀疏、③难以标准化回测的结构性问题。从交易范式看，2026 年的宏观环境更接近 1999-style regime：①高估值科技叙事、②地缘政治不确定性抬升、③真实经济基本面的反复验证三条主线相互缠绕。在这一框架下，AI Tagging 通过对①叙事扩散、②风险因子暴露、③资产联动的持续标注，为 macro hedging alpha 提供了更稳定的信号层。

③ Validating execution at the trading desk level.

At the execution layer, a futures paper-trading pipeline is built in a testnet environment to test (1) matching logic, (2) slippage assumptions, and (3) risk-control parameters. This environment is chosen because professional market-making (MM) teams are already testing strategies on the same testnet, allowing liquidity, order-book behavior, and counterparty actions to be observed more realistically.

在交易桌执行层面，基於 testnet 环境搭建 futures paper trading 流水线，用於验证①撮合逻辑、②滑点假设、③风控参数。之所以选择该环境，是因为已有专业做市商（MM）团队在同一测试网络进行策略试盘，能更真实地暴露流动性、盘口与对手方行为。

Stories of how macro traders earn in the market?

AWS FSI Partner Sponsors (Golden Tier)

① Sensors Data (神策数据)

Banks are divided into different types: retail banking, investment banking, virtual banking, and commercial banking. Because of this, a full 360‑degree customer journey looks different for each type. For example, when activating deposits, AI must understand user experience and financial product promotion in different ways. By using each bank’s own business data, AI helps match products to real market needs.

银行细分成「零售银行，投资银行，虚拟银行，商业银行」。因此，360客户日程各有异同，比如盘活存款时，AI理解用户体验和推销理财产品的过程都不一样，通过公司的业务数据来契合市场需求。

② IntSig (合合信息)

IntSig’s strength is long‑context AI reasoning. In finance, balance sheets span long time periods and often require reading rolling annual reports to understand the business value behind each number. With long‑context AI, IntSig allows AI to directly understand financial accounting logic, turning complex financial data into real business actions.

超长上下文推理是合合的AI优势。金融行业的资产负债表是超长距度，需结合「滚动年报」才能解读每一个数据背後的商业价值。而合合信息利用AI超长上下文，让AI直接拥有「金融会计」的数字感知，交付真正读懂金融数据的商业行动。

③ InsureMO (易保云)

InsureMO has built a full set of insurance‑focused, “lobster‑style” APIs that are highly practical. These APIs give AI strong built‑in capabilities, including customer KYC, credit checks, background audits, and message delivery across front, middle, and back offices. This helps insurance AI fit market needs more directly.

沉淀了保险行业的龙虾型API最适用服务，等於直接提供AI最全面的power skill，包括由前中後台的客户KYC，信贷核查，背调审计，发送短信结果等，将保险龙虾AI契合市场需求。

④ DBAPPSecurity (安恒信息)

Its security engine is built on 20 years of cybersecurity experience, using rule libraries developed through real red‑team penetration testing. This kind of human security judgment cannot be replaced by AI. By bringing human security thinking into business operations, AI can help monitor and control security risks at the business level. This represents a new future model for enterprise security.

安全引擎是来自二十年资安经验，由红队渗透而建立的安全规则库，这是AI无法取代的人类思捷。同时，把「人类安全思捷」引入到「商业行动」，AI在业务层把关商业市场的安全风险，这是未来的安全新范式。

⑤ Deloitte Consulting (德勤咨询)

Deloitte’s AI advantage lies in topology relationship graphs. By combining large client data sets with mathematical algorithms, AI can identify patterns and connections more reliably. This creates new insights for the consulting industry and delivers real business value to top 100 enterprises.

拓扑关系图是德勤的AI优势，结合了庞大的客户库数据，通过数学算法让AI稳定识别信息，这是咨询行业的新洞察，爲百强企业提供商业行动价值。

My name is Kenny Chan. I work as a macro trader and structured broker, and I also use QuantAI tools as part of the AWS Community Builder.

I focus on global markets, manage profits, and aim to build a strong career in trading. My main skills include macro trading, structured brokerage, AI-based solutions, and risk management.

We are an innovative AI-based financial services company focused on technology-driven trading. We also explore areas of tech innovation, such as AI venture capital and private equity, macro and crypto quantitative trading, and CTF financial security compliance.

我们是一家创新型AI金融服务企业，专注於技术驱动交易。更涉猎科技创新领域，如AI创投私募，宏观和Crypto量化交易，CTF金融安全合规等。

Portfolio Management for Macro Hedging (Updated:2026-01-09)

Kenny Chan — Fri, 21 Nov 2025 05:19:27 +0000

宏觀對沖的組合管理 Portfolio Management for Macro Hedging

核心觀點

第二代宏觀對沖，包含三個部份：宏觀面、技術面、交易系統。但2025年有三大痛點：①第一，遲疑了未戰略轉移到權益資產，錯過了2025年的中歐美權益共舞。②第二，遲疑原因，是未懂如何去回測宏觀對沖的几何收益路徑。③第三，因此未知風險的恐懼令我做錯了，即使2025年是實現每月正收益。
如何破局？我要避免陷入半夏宏觀的“忘記了自己的看家本領”，同時提升宏觀對沖的組合管理工藝。因此，我定了兩個KPI指標：①每月正收益實現率，Pure-Alpha絕對淨值。②Sharpe vs Sortino，宏觀對沖下跌風險的工藝。
第三代宏觀對沖系統，在現有的架構上升級了：①Quarterly Correction Assets，強化宏觀叙事面的戰略轉移。②90-Day Rolling Window Correction Assets，強化Sharpe vs Sortino 即提減低風險同時提升收益率。③蒙地卡羅回測宏觀對沖的組合管理，即几何收益路徑來決定“風險+收益率”表現。

Core Idea

The second generation of macro hedging has three parts: macro view, technical view, and trading system. But in 2025, there are three big problems: ①I hesitated to shift strategy to equity assets and missed the U.S.–Europe equity rally in 2025. ②The hesitation came from not knowing how to backtest the geometric return path of macro hedging. ③Fear of unknown risks led to mistakes, even though 2025 delivered positive monthly returns.
How to Break Through? I must avoid falling into the trap of “forgetting my core skill” and improve portfolio management for macro hedging. So, I set two KPIs: ①Monthly Positive Return Rate and Pure Alpha Net Value. ②Sharpe vs Sortino to refine risk control in macro hedging.
Third-Generation Macro Hedging System. The upgraded system adds three key features: ①Quarterly Correction Assets – strengthen strategic shifts based on macro narratives. ②90-Day Rolling Window Correction Assets – improve Sharpe vs Sortino to reduce risk and boost returns. ③Monte Carlo Backtest for Macro Hedging Portfolio Management – simulate geometric return paths to measure combined “risk + return” performance.

1.0 Quarterly Correction Assets

強化戰略轉移。負相關是Macro Hedging的核心。如果所有資產都正相關，即黑天鵝危機全部市場下跌。
Strengthen Strategic Shifts. Negative correlation is the core of macro hedging. If all assets move together, a black swan event will crash every market.
所有資產趨正相關時，宏觀對沖表現不佳。由2020-Q1至2025-Q4，有5個季度出現所有資產趨正相關。其中3次是所有資產下跌（2020-Q2、2022-Q2、2024-Q1），宏觀對沖的大盤同步大幅回撤，即宏觀對沖失去了分散風險的效果。其中2之次是所有資產上升（2025-Q3、2025-Q4），宏觀對沖收益率跑輸大盤。在近8年後疫情大動蕩，宏觀對沖失去了分散風險的效果，收益率也跑輸大盤，如果只是每月正收益但續效不甚理想。
剖析過往5次所有資產趨正相關。①一致性下跌：2020-Q2是疫情之始環球量化寬松。2022-Q2是量化寬松副作用+中歐美經濟不及預期。2024-Q1是高息周期+經濟修復觀察。②一致性上升：2025-Q3和2025-Q4是環球AI共舞+中歐美經濟熱絡。
因此，宏觀收益流模型用來強化戰略轉移。因每個資產的質地隋時間變化，我們利用宏觀收益流模型，提取每個資產當刻的三個關鍵收益流，分隔開組合的風險和收益，從而強化戰略轉移的敏捷度。

✅2020-Q2 美債：安全港、美元信用、疫情
✅2020-Q2 黃金：安全港、通用貨幣、疫情
❌2022-Q2 美債：加息周期、滯脹（通脹+經濟不及預期）
✅2022-Q2 黃金：通用貨幣、對沖滯脹（通脹+經濟不及預期）
✅2024-Q1 商品：經濟修復觀察
❌2025-Q3 商品：中歐美經濟熱絡

提取每個資產的質地。①在2020-Q2美債和黃金是“安全港”，但到2022-Q2是量化寬松副作用，“通脹”主線下，對美債是加息風險，而黃金則是收益良機。②在2024-Q1是“經濟修復觀察”，利好商品交易，因供需雙邊彈性都大，但到2025-Q3“中歐美經濟熱絡”反而是利空商品交易，因壓縮了供需雙邊彈性。
潛在風險：所有資產趨正相關。①一致性下跌，如環球疫情、地緣戰爭、國家危機。②一致性上升，如環球AI共舞，中美歐經濟需求熱。

2.0 90-Day Rolling Window Correction Assets

減低風險同時提升收益率。通過時間可視化 Sharpe波動率 vs Sortino下跌風險。
Lower Risk and Raise Returns. Use time-based visualization of Sharpe (volatility) vs Sortino (downside risk).
把時間當成朋友，而非敵人。長遠地看哪些資產在近三十年，哪些資產組合最有效率地對沖。
首先，90天中周期合適做月度正收益，而240天長周期合適做超額收益。90天中周期波動率在+-10%，而240天長周期波動率在+-30%。但頻率上看，90天中周期有2~3次+10%收益機會在一年內，但240天長周期只有1次+20%收益機會在一年內，但伴隨著更大的虧損風險。因此，我會選擇90天中周期來做基線，來實現每月正收益宏觀對沖。

❌2022-03 HSI恆生指數 長周期（-37%） 中周期（-17%）
❌2022-10 HSI恆生指數 長周期（-41%） 中周期（-23%）
❌2024-01 HSI恆生指數 長周期（-30%） 中周期（-11%）
❓2024-11 HSI恆生指數 長周期（+10%） 中周期（-14%）
✅2025-03 HSI恆生指數 長周期（+46%） 中周期（+23%）

其次，長周期只是能回本，但虧損風險和中周期是一樣。如近5年的HSI恆生指數，長周期和中周期同樣遭遇3次熊市（2022-03、2022-10、2024-01），因長周期的3次虧損更深，最終只能回本。只有一次是例外，2024-11長周期的第一波牛市回調，但只要堅持到2025-03第二波牛市，兩者皆能盈利，而長周期跑出一倍收益。因此，我會選擇90天中周期來做基線，把風險虧損控制在10%以內。
潛在風險：180日以上的長周期趨勢。①長牛市，跑出一倍以上的超額收益。②長熊市，，跑出一倍以上的超額虧損。

3.0 蒙地卡羅回測宏觀對沖的組合管理 Monte Carlo Backtest for Macro Hedging Portfolio Management

几何收益路徑才是首要關注，而不是平均收益。
Focus on Geometric Return Path. The priority is the return path, not the average return.
通過蒙地卡羅回測200萬次几何收益路徑，找出5%~95%置信區，剖析了各種不同風險的形態，同時減少了未知的恐懼壓力。
幾何算術vs平均算術。簡單來說，幾何算術是單獨計算每個獨立字宙的結果，然後觀察最好&最差的字宙結果。而平均算術是總納了所有獨立字宙的結果。但在交易世界，人每次只能選擇一次，人生路是亦因亦果，因此，要用幾何算術，勇敢地面對過往的成敗，而不能用“假設”的其他字宙平均期望值。
宏觀對沖目標不是跑嬴大盤，而是細水長流，穩穩地踏實做交易。以NAV=100來計算，最好的宏觀對沖組合，Q50是103，Q75是108；而SP500的Q50是103，Q75是113；即是在多個宇宙SP500的收益率，都優於宏觀對沖組合。但以風險來看，SP500的Q75是93，Q5是85；而宏觀對沖組合Q75是98，Q5是83；即是在多個宇宙宏觀對沖組合的下跌風險，都優於SP500。因此，我會選擇宏觀對沖，細水長流。
200萬次Block Bootstrap Monte Carlo。與上一代的回測比較，今次更著重了宏觀對沖的兩大看家本領“擇時+配置”。因每個時間點上，風險形態和收益質地都不一樣，所以主動選擇90days風險曝光，來量度各時間點的宏觀敍事下，風險收益的形態質地。而上一代回測，只著重“倉位，買賣，資產”這三大支柱。
潛在風險：黑天鵝危机。進入了5%置信區，蒙地卡羅模型失效。

4.0 宏觀對沖實務 Macro Hedging in Practice

每個“資產、時間、收益、風險”皆不平等。有些資產，天生是“時間+收益”的朋友，“風險”的敵人。
Every asset, time, return, and risk is unequal. Some assets are natural friends of time + return and enemies of risk.

4.1 第一不平等：沒效率的對沖 First Inequality: Inefficient Hedging

“富時A50指數+中證商品指數+黃金”是低效的對沖資產。因中國是以經濟政策市主導，二者皆有錯位順周期的屬性。當強政策出台，三個資產趨正相關，也導致了中國宏觀交易表現不佳。xin9-ccicfi-gld，90days-q50是101.7，270days-q50是105.5；對比hsi-tlt-gld 90days-q50是102.2，270days-q50是106.5。

4.2 第二不平等：時間等於風險 Second Inequality: Time Equals Risk

“恆生指數+二十年美債+黃金”的時間等於風險。因恆生指數與上證指數相關性高，收益流同是“經濟政策市+非美元資產”主導，所以“美債+黃金”不能對沖背後的收益流，因此增加時間曝光提高了風險，但未提升收益。hsi-tlt-gld，90days-q25是97.8，而270days-q25是98.5；90days-q5是91.3，而270days-q5是87.7。

4.3 “擇時+配置”兩大核心，在美國市場既有效率對沖，也有時間收益，而減少風險。 Timing + Allocation — make U.S. markets efficient for hedging, while also giving time-based returns and reducing risk.

“長慢牛，短快熊”是SPX的時間收益，加上“美債+黃金”有效對沖SPX的收益流。spx-tlt-gld，90days-q5是93.9，270days-q5是92.9，下跌風險明顯降低，較前兩個組合。而spx-tlt-gld，90days-q25是99.5，270days-q25是102.2，即增長時間曝光可提升收益。

4.4 每個“資產、時間、收益、風險”皆不平等。 Every asset, time, return, and risk is unequal.

從資產類別看，美股是“長慢牛，短快熊”。A股港股是“短快牛，長慢熊”，債劵匯率是“慢上慢落，小概率突跌空插水”，商品是“慢慢地上下樓梯”。從時間看，權益類資產是“3~4年牛熊周期”，債劵匯率是“5~8年牛熊周期”，商品是“10~20年牛熊周期”。
因此，每位交易者從不同的視角入手，只做自己最擅長的“收益風險形態”。行業共同基金是基於“資產”賺質地，指數基金是基於“資產+時間”賺國運，橋水全天侯是基於“風險+時間”賺穿越周期，耶魯組合是基於“風險單位加杆杠”賺風險本身。
我最擅長是“擇時+配置”，所以會選擇橋水全天侯的基於“風險+時間”賺穿越周期。

以A50從2021年至2025年，90days是“牛多熊少”，而270days是“牛少熊多”。因此，在同一資產但不同時間框架看，90days是有“時間+收益”且“減少風險”，對是“擇時+配置”最有利。90days的>10%牛市次數有4次，高於270days的>20%牛市次數2次。90days的>-10%熊市次數有3次，低於270days的>-10%熊市次數4次。

4.5 三大支柱：宏觀面、技術面、交易系統。蒙地卡羅几何收益路徑優化“宏觀面”。Three pillars: macro view, technical view, and trading system. Use Monte Carlo simulation to optimize geometric return paths for the macro side.

①宏觀面（宏觀收益流模型、季度負相關戰略、宏觀對沖配置）
②技術面（支持位，阻力位）
③交易系統（倉位、買入、止盈、止損）

蒙地卡羅几何收益路徑+宏觀對沖配置。

第一，界定下限：預計每次最大風險是-10%，通過90days-q5置信區壓力測試。
第二，界定上限：預計每次最大收益是+10%，每年有2~3次机會，通過q25~q50置信區壓力測試。
第三，強化戰略轉移+宏觀對沖配置：好的戰略轉移是增加收益且減少風險。
第四，資產+時間皆不平等：SP500天生質地好，A50做波段天生好。

4.6 杠杆+複利，是宏觀對沖的兩大隱藏優勢。Leverage and compounding are two hidden strengths of macro hedging.

宏觀對沖的不是跑嬴大盤，而是要穩定的收益，即交易時的幸福感。而這幸福感，即與時間做朋友，就擁有兩隱藏優勢。

杠杆。如果最大回撤是10%，即可用10x杠杆才會爆倉歸0。而純股票多頭不能用杠杆。
複利。時間愈長收益愈多，人生更安全。而純股票多頭只要做錯，跌25%就要花4倍時間追平。

4.7 宏觀對沖交易實務。Macro hedging in practice.

第一部份，目標
①人生目標：有什麼夢想？
②交易優勢：比其他交易者有什麼優勢？
③虧損下限：最大損失多少錢？
④盈利上限：要賺多少錢才能滿足夢想？
第二部份，資產
①點算資產：負債多少？固定收入多少？彈性收入多少？
②計算開支：固定開支多少？日常消費多少？
③預計成本：用多少資產做交易？
④分配時間：每天交易8小時自營交易員？
第三部份，紀律
①設計架構：個人品牌是什麼？交易風格是什麼？
②撰寫筆記：覆盤自己的強弱項？
③創意擇優：選擇最好的東西，而不是最喜歡的東西？
④執行高效：選擇最高價值的路，而不是最舒服的路？
第四部份，進步
①極端開放：如何學習？閱讀？討論？
②目標導向：有意義的夢想？
③擁抱失敗：輸在哪？錯在哪？
④優化系統：為什麼？下一步如何做？
第五部份，交易
①資產池：高流動性+質地好。交易所上市大盤指數ETF & OTC場外基金。
②宏觀面：宏觀收益流模型、季度負相關戰略、宏觀對沖配置。
③技術面：支持位，阻力位、1|3|5|10年高低位、三天法則。
④交易系統：倉位、買入、止盈、止損。

5.0 交易執行實務。①目標、②資產、③紀律、④進步、⑤交易。Trading execution essentials: ① Goals, ② Assets, ③ Discipline, ④ Growth, ⑤ Trades.

5.1 資產池 Asset Pool

16個不同資產，多元化+高流動性+質地好+低費率。
$SPX,$COMPQ,$HSI,$HSCEI,$SSEC,FXI,TLT,$BRENT,$CRB,$USD,$CNYUSD,GLD
只有MM流動性的上市產品，有①對手風險和②負溢價。即只能跟MM做對手盤。
高流動性的產品，足夠深也足夠多交易者，才能防止①對手風險和②正溢價。
1個月內高成交額，而非高成交量。高成交額的產品，即高流動性，有利大資金做中長線部署。但高成交量產品，不代表高流動性，只有利做場內套利或短線算法交易。高流動性產品有02800.HK/02828.HK/03033.HK等大盤指數。
高成交量不代表高流動性，07552.HK/07226.HK/07500.HK等只是+2x/-2x等磨損極高的每日追蹤ETF。

Symbol	Tracking Index	Turnover(Price * Vol, 1M)	Listed Year
02800.HK	Hang Seng Index	259.38B HKD	1999
02840.HK	LBMA Gold Price PM ($/ozt)	1.72B HKD	2014
03086.HK	NASDAQ 100 Index	104.35M HKD	2018
SPY	S&P500	978.02B USD	1993
QQQ	NASDAQ 100 Index	668.59B USD	1999
GLD	LBMA Gold Price PM ($/ozt)	69.37B USD	2004
TLT	US Treasury 20+ Year Index	45.14B USD	2002
USO	Front Month Light Sweet Crude Oil	4.88B USD	2006

5.2 宏觀面 Macro Analysis

通過指數化交易，第一，有效地用40+年歷史數據回測。2016~2025年組合年度最大回撤是17%~25%，而收益率90days是+8%，270days是15%~25%。但這些只是從年化的視角上看，而不是真正的交易實務上看。因此，後續需用蒙地卡羅驗證。
只有少量的多元化配置是不足夠。只有分別25%的股債商金，才可以實現低回撤。（回測數據由2016年～2025年）
Macro Hedging Portfolio (Equal) 25% QQQ, 25% TLT, 25% USO, 25% GLD
Annualized Return: 10.5% ⭐
Standard Deviation: 10.8%
Max Drawdown: 17.6% ⭐
Macro Hedging Portfolio (Risk-offset) 40% QQQ, 40% TLT, 10% USO, 10% GLD
Annualized Return: 10.5%
Standard Deviation: 10.8%
Max Drawdown: 23.9%
Macro Hedging Portfolio (Alternative) 10% QQQ, 40% TLT, 25% USO, 25% GLD
Annualized Return: 7.2%
Standard Deviation: 9.9%
Max Drawdown: 17.0% ⭐
用交易實務視角，再跑蒙地卡羅回測，得出之前相同的驗證。90days短期看，Alternative更受商品10+年長周期而看似更優。但270days長期看，Equal更有細水長流的宏觀對沖優勢。因此，我會選擇各25%的Equal，時間上彈性地由90days曝險至270days。
2002~2025, 90days
Portfolio_Equal - Win%: 72.06%
Portfolio_RiskOffset - Win%: 75.17% ⭐
Portfolio_Alternative - Win%: 69.82%
2002~2025, 270days
Portfolio_Equal - Win%: 81.48% ⭐⭐⭐
Portfolio_RiskOffset - Win%: 85.44%
Portfolio_Alternative - Win%: 79.57%
蒙地卡羅回測，是几何算術收益路徑，即在實際的交易實務更優，較用平均算術的組合管理。因此，對交易者來說，蒙地卡羅回測更能幫助交易的執行盤感。
本土情結是交易者大忌，不能帶著情緒來做交易，“喜歡vs最好”是不一樣。用02888.HK取代QQQ，win%下跌8%，q95和q75分別下跌2%。有兩個原因，第一，恆生指數對沖不夠高效率，有時間滯後性。第二，恆生指數質地不夠好，沒時間收益，長熊短牛。
Portfolio_Equal - Win%: 65.67%
ortfolio_RiskOffset - Win%: 64.10%
Portfolio_Alternative - Win%: 67.57%

第二，可低成本地複制組合，比如買石油股共同基金貝萊德世界能源基金，大体上可追蹤WTI原油期貨。
①組合設計：資產、時間、收益、風險 ②交易執行：倉位、買入、止盈、止損
第三，指數分為兩種，①直接交易指數，②輔助組合指數。如美元指數是輔助指數，用來幫助宏觀收益流模型。如美元指數升，會有美元溢價影響股債油金匯。

交易指數	品種	地區
S&P500 Index	Stock	US
NASDAQ 100 Index	Stock	US
US Treasury 20+ Year Index	Bond	US
S&P 500 Energy Index	Oil	US
LBMA Gold Price PM ($/ozt)	Gold	Global
Dim Sum Bond Index	Bond	HK
Hang Seng Index	Stock	HK

輔助指數	品種	地區
A50 Index	Stock	China
Euro Stoxx 50 Index	Stock	Euro
USD Index	Fx	US
RMB Index	Fx	China
BTC Index	Crypto	Global
WTI Oil Index	Oil	Global

交易執行，計算最便宜的匯率表和兌換時機。Trading means picking the cheapest exchange rate and the right time to swap money.
2025/11/25 HKDUSD 是7.77818，是中立偏高水平，但相較2022年~2024年強美元時期，目前美元定價已不貴。隨著美聯署2025年減息，美元定價會再有下跌空間至7.76，但不會跌至7.75的1993~1999年克林頓大穩定時期和2020年的疫情沖擊。

PERCENTILE	HKDUSD
Q0	7.7056
Q5	7.731
Q25	7.7506
Q50	7.7681
Q75	7.7994
Q95	7.8458
Q100	7.8503

1993/01/01~2025/11/25。近32年HKDUSD匯率，中位數（中性美元）是7.7681，上四分位數（貴美元）是7.7994，下四分位數（弱美元）是7.7506。極端貴美元（Q95）是7.8458，極端弱美元（Q5）是7.731。

11/27/2025	Customer Sell	Customer Buy
券商A	7.7655	7.7895
銀行B	7.7531	7.8033
銀行C	7.7472	7.8088

券商A對交易者最友好。買入美金是最低價，賣出美金最高價，較香港另外兩間銀行（銀行B，銀行C）。

券商A Customer Buy收取100滴(+0.01224)的價差。HKDUSD是7.77731，券商A Customer Buy是7.78955，相差+0.01224。
因此，交易者在券商A上兌換即可。但要留意①美元水平，②美元方向。另外，③減少磨損，只單邊兌換，而不要來回HKD->USD兌換導致資本磨損。

HKDUSD最大匯率波幅是2.00%，即1Y無風險利率~3%可回補匯率貶值損失。
2025/12/03 HKDUSD是7.7835。近30年歷史數據顯示，最多向下貶值-1.00%，最多向上升但0.86%。

	情景	HKDUSD 牌價	HKDUSD 變化%	美元年化收益%	美元年化收益% (匯率曝光)	港元年化收益%	美元 vs 港元年化收益%(匯率曝光)
最好情景	美元指數升值>$105	7.7994	0.2043	4.2000	4.4043	3.1000	1.3043
中性情景	美元指數維持$95~$105	7.7835	0.0000	3.7278	3.7278	2.4000	1.3278
最差情景	美元指數眨值<$95	7.7506	-0.4227	3.3000	2.8773	1.9000	0.9773

在任何情景，美元收益率都比港元收益率高0.93%(最差情景美元貶值)~1.32%(最好情景美元升值)。
第一，聯繫匯率下近30年最大波幅只有2%，現金寶年化收益率可抵消匯率曝光。第二，美元天生是優質資產，現金寶年化收益率較高。

中性至最好場景(美元升值/美元維持不變)下投資美元是最優，比港元多賺1.30%年化收益率。而最差情景(美元眨值)下，只比港元多賺0.97%年化收益率，因美元眨值損害了收益率。

2024~2025年美聯署減息公布，對USDHKD聯繫匯率沒直接關系。

日期	減息幅度	備註
2024年9月18日	–50 bps	首次 4 年內降息，由 5.25–5.50% 至 4.75–5.00%
2024年11月7日	–25 bps	降至 4.50–4.75%
2024年12月18日	–25 bps	第三次降息，降至 4.25–4.50%
2025年9月17日	–25 bps	首次減息至 4.00–4.25%
2025年10月29日	–25 bps	再降至 3.75–4.00%

兌換USDHKD時機基於 ①港元資金結構 + ②美聯署公布情緒預期。第一，USDHKD的影響因素是來自港元資金結構，如拆息等。第二，USDHKD已消化了美聯署減息公布的情緒預期，USDHKD是非常有效率的市場。
2025年12月05日，港元USDHKD定價不算貴($7.7835，在Q50中位數上方，但比2025/05~2025/08的$7.85高位已回落-0.84%)，另外港元資金結構平穩(3個月HIBOR是3.0082%)。

券商A兌換況程，所有流程需在九點前HKT操作，即T+1。

Datetime	Action
2025/12/16 08:50AM	贖回100,000.00HKD。
2025/12/16 17:00PM	T+1到帳100,000.00HKD，早上九點前HKT贖回。
2025/12/17 09:13AM	申請兌換。HKD->USD。總共100,000.00HKD，USDHKD牌價是0.12834。
2025/12/17 09:16AM	USD到帳。兌入12,834.32USD。
2025/12/17 09:25AM	申講12,833.32USD。
2025/12/18 17:00PM	T+1收益。早上九點後HKT申講。

宏觀收益流模型，是交易資產背後的驅動因素，即人是如何看這個世界的脈絡。A macro model shows what makes people trade — basically, how they think the world is going.

宏觀六大變量：

經濟增長與週期（GDP、週期、消費信心）
貨幣與利率政策（央行政策、利率、實質利率、通膨、QE）
匯率與美元相關（美元／人民幣強弱、指數趨勢、利差）
地緣政治與風險（戰爭、信用風險、資本外流等）
行業與技術創新（科技創新、能源／商品供需、產業採用）
資本流動與避險需求（全球流動性、避險需求、貿易收支）

1. 經濟增長與週期
GDP 成長與經濟週期：如 #us-gdp-growth、#china-gdp-growth、#economic-growth-cycle。
消費與信心：如 #consumer-confidence。

2. 貨幣與利率政策
央行政策與利率：如 #fed-rate-policy、#ecb-monetary-policy、#pboc-fx-policy。
量化寬鬆與流動性：如 #fed-qe-policy、#global-liquidity。
實質利率與通膨預期：如 #real-interest-rates、#inflation-expectations。

3. 匯率與美元相關
美元強弱與匯率趨勢：如 #usd-strength、#usd-index-trend、#usdcny-rate-spread。
人民幣與流動性：如 #rmb-fx-trend、#rmb-liquidity。

4. 地緣政治與風險
戰爭與地緣風險：如 #geopolitical-risk、#geopolitical-capital-flow。
監管政策與制度風險：如 #regulatory-policy。

5. 行業與技術創新
科技創新週期：如 #tech-innovation-cycle。
能源與商品供需：如 #global-supply-demand、#energy-inflation-pressure。

6. 資本流動與避險需求
避險需求：如 #safe-haven-demand。
資本流動與信用風險：如 #china-credit-risk、#trade-balance。

Index	Macro Tags	Impact Direction
S&P 500 Index	#us-gdp-growth, #fed-rate-policy, #inflation-labor-market	GDP↑ → 利多; 升息 → 利空; 高通膨 → 利空
NASDAQ 100 Index	#tech-innovation-cycle, #interest-rate-sensitivity, #usd-strength	科技創新↑ → 利多; 利率↑ → 利空; 美元強 → 利空
Hang Seng Index	#china-policy-support, #rmb-fx-trend, #consumer-confidence	政策寬鬆 → 利多; 人民幣強 → 利多; 信心弱 → 利空
A50 Index	#china-gdp-growth, #policy-easing-measures, #geopolitical-capital-flow	GDP↑ → 利多; 降準降息 → 利多; 地緣風險↑ → 利空
LBMA Gold Price PM	#real-interest-rates, #usd-index-trend, #geopolitical-risk	實質利率↓ → 利多; 美元強 → 利空; 地緣風險↑ → 利多
US Treasury 20+ Year	#inflation-expectations, #fed-qe-policy, #safe-haven-demand	通膨↑ → 利空; QE → 利多; 避險需求↑ → 利多
WTI Oil Index	#global-supply-demand, #economic-growth-cycle, #geopolitical-risk	需求↑ → 利多; 經濟強 → 利多; 戰爭風險↑ → 利多
Dim Sum Bond Index	#rmb-liquidity, #china-credit-risk, #global-rate-environment	RMB流動性↑ → 利多; 信用風險↑ → 利空; 全球利率↓ → 利多
Euro Stoxx 50 Index	#ecb-monetary-policy, #eurozone-economic-growth, #energy-inflation-pressure	ECB寬鬆 → 利多; 經濟強 → 利多; 能源通膨↑ → 利空
USD Index	#us-interest-rates, #safe-haven-demand, #trade-balance	利率↑ → 利多; 避險需求↑ → 利多; 貿易逆差↑ → 利空
RMB Index	#china-export-performance, #pboc-fx-policy, #usdcny-rate-spread	出口強 → 利多; 央行維穩 → 利多; 中美利差擴大 → 利空
BTC Index	#global-liquidity, #regulatory-policy, #crypto-adoption-rate	流動性↑ → 利多; 監管嚴 → 利空; 採用率↑ → 利多

宏觀對沖已是“交易匯率”。因此，“匯率溢價”應計算在總收益或總虧損，而不需鎖匯對沖“匯率風險”。Macro hedging already includes currency moves. So, currency gains or losses should be counted in total return, without locking the exchange rate to hedge currency risk.
案例1：中資美元債有三個宏觀收益流變量：①利率水平的鬆緊邊際變化，②中國企業的信用質地，③美元流動溢價的鬆緊邊際變化。因此，美元的升跌方向，也是收益的一部份。如看跌美元，即看跌中資美元債，而不該鎖匯對沖。
案例2：美股和美元的相關性不明顯，因此宏觀對沖把“美元”看成一個交易變量，而非對沖的風險。最佳情景是“賺美元+賺美股”。次佳情景是“虧美元+賺美股”。最差情況是“虧美元+虧美股”。

5.3 技術面 Technical Analysis

①歷史感，知高低。1|3|5|10年高低位。
②價格感，知分寸。支持位、阻力位。
③順大勢逆小勢。三天法則。

5.4 交易系統：倉位 Trading System: Position Sizing

①組合、②時間、③收益、④風險、⑤本金。用任何一點，都可求出最優的答案。

通過風險單位量度交易本金。如最大虧損$800USD，組合A最大回撤5%，可用$16000USD進行交易。
通過交易本金量度交易組合。如可用$16000USD，最大回撤5%，可交易組合A進行交易。
通過交易組合量度交易本金。如交易組合A，最大回撤5%，最大盈利10%，可用$16000USD進行交易。
通過時間單位量度風險單位。如90days，交易組合A最大回撤5%，最大盈利10%，可用$16000USD進行交易。

5.5 交易系統：買入、止盈、止損 Trading System: Entry, Take-Profit, Stop-Loss

（略。）

6.0 宏觀交易機構跟別人不同。我們聚焦在“交易風險形態”。Macro trading firms are different. We focus on the shape of trading risk.

6.1 宏觀交易員SOP3.5

SOP 3.0 vs 3.5的異同，每一步都是人機協同，只是把AI自動化比例提高，釋放生產力。宏觀對沖交易實務的七步，v3.0是把其中兩步100%自動化，但這是錯誤的優化。因此，v3.5把每一步的AI自動化比例提高，幫助交易者更聚焦人的價值。最後是工作時間減少了，而生產力提升了。

宏觀對沖交易實務的七步。每一步都重要，但時間占比%各不相同。把每一步的AI自動化比例提高，幫助交易者更聚焦人的價值。AI自動化占60%，手動占40%。人類工作時間減少了，而生產力提升了，通過AI自動化。

6.2 宏觀交易員SOP3.5、輸入、輸出、可控指标KPI

步驟	輸入	輸出	可控指標KPI
1. 風險	資產池、標的名單	最大虧損/收益、90/180/270天滾動波動率	資產池多元化>5種、流動性>1B/月、總費率<0.9%
2. 時間	窗口長度	負相關驗證	資產間相關<-0.1、90天下跌風險<-5%、正相關熔斷>0.75
3. 組合	標的流動性、交易費率	置信區分位數、目標風險配置、月度盈利%	Q5路徑>-10%、每月正收益>90%、90基線vs270超額<10%
4. 宏觀面	宏觀事件	季度策略、配置建議	90天基線一致[組合]、敘事一致性、季度更正
5. 技術面	價格、週期	支撐阻力區、入場/離場訊號	入離場近支撐/阻力區、三天法則右側交易
6. 倉位	最大允許虧損、回撤/止盈目標、資產波動率	組合大小、每筆大小	每筆風險<2%本金、風險/收益一致[宏觀面]
7. 進出場	風險限額、止盈目標、入場/離場條件	盈虧、交易執行	止盈≈+20%、止損≈-5%、執行一致[宏觀面][技術面][倉位]

6.3 宏觀交易員SOP3.5、AI自動化工具、交易員工具、總占比%、自動%、手動%

步驟	AI自動化工具	交易員工具	總占比%	自動%	手動%
1. 風險	滾動窗口波動率	Choice	10	80	20
2. 時間	負相關圖	Choice	10	50	50
3. 組合	蒙地卡羅回測	portfoliovisualizer	15	55	45
4. 宏觀面	宏觀收益流模型	InsightReport	30	45	55
5. 技術面	Backtrader	TradingView	10	70	30
6. 倉位	Backtrader	Excel戰略筆記	15	80	20
7. 進出場	Backtrader	Futunn	10	60	40

Trading Backward交易逆向法。從現有的資產配置，實現戰略轉移，需從最後的交易執行開始。第一，倉位大於技術面，倉位是基於技術形態而微調。第二，宏觀面大於時間和組合，由宏觀來決定持有什麼資產和時限。因此，Trading Backward交易逆向法是由“交易+倉位+宏觀+風險”四個關鍵部件驅動。雖然Trading Backward最近距離地看見收益，但只能通過以上四個關鍵部件來操作。

With QuantAI and AWS, I built a third-generation macro hedge system focused on Multi Asset Timing and Alternative Asset Allocation. In a highly volatile market, this system delivered 12 months of positive returns.

基於QuantAI與AWS，我構建了第三代宏觀對衝系統，專注於多資產擇時與另類資產配置，在高波動市場中實現12個月正收益。

Kenny's Trading Journey 2026 into Alternative Investments

Working together on the trading desk to create better Alpha. Using big-picture research from macro trading to help crypto trades run faster and more efficiently with high-frequency short-term strategies.

交易桌的協同Alpha 。宏觀交易的大勢研判投研，幫助Crypto交易做更有效率的HFT短線交易。

7.0 交易以外的工作

對外有：①戰略執行、②演講路演；對內有：③跨領域協調、④AI賦能、⑤成本收益比率。

7.1.1 個人品牌Pitching。量化AI、宏觀交易員、前台交易桌。QuantAI、Macro Trader、Broker 。

掌控全球宏觀利潤，運用量化人工智能，打造精英交易生涯。
Control global macro profits, use QuantAI, and build an elite trading career.

個人品牌戰略，是“企業+個人”的雙向成長飛輪。因此，個人品牌戰略，更強調“中長期的隱性價值”，且定位是更“專業性+差異化”，有時間有土壤培養，而不是自媒体網紅的短期流量變現。

對企業中長期的隱性價值：“專業性+差異化”個人品牌戰略，提供了“信任溢價&服務溢價”，降低了“不被信任的磨合成本”。更理想是精品專業化，帶來了“更高利潤的高端品牌市場”。

對個人的隱性價值：通過企業飛輪，個人也得到“社交資本”。從內部看，是“行業資源+崗位機會”。從外部看，是“思想影響力+職場提升機會”。

7.1.2 個人品牌分成三個階後：定位，成長，精品

第一，定位階後。從自己出發，考慮自己的Tag關鍵字。從時間，地點，職業，興趣，夢想，價值觀等。
#香港 #港城大 #交易 #創投 #藝術 #技術 #寫作 #演講 #音樂 #電影 #幸福感

第二，成長階後。專業化+差異化，通過兩種微創新方法。①大路中的小路，②新方法融合。
#宏觀交易者 #Broker #QuantAI

第三，精品階後。輸出行業深度，戰略價值，跨領域協調。
#蒙地卡羅的組合管理 #Portfolio-Management-for-Macro-Hedging

7.1.3 我的品牌矩陣。IC精品行業家，提供戰略價值，跨領域協調，且有專業技術深度。走輕管理風格。

LinkedIn：職場門面。戰略價值，跨領域協調。
Medium：深度長文。專業技術，外部影響力。
Twitter：即時的行業想法。職場里的非重要事。

7.1.4 從人的角度去想，他是如何想，如何做。

建立個人品牌，和跨領域協調，過程中皆是人構成。而過程中有很多隱性價值，比如人的喜好，情緒，名譽等等，導致爭吵阻力。而這些阻力不能運用基本邏輯解釋。

因此，要先了解對方如何想，如何做，才可以通過Leadership制定下一步策略。

懶慵慵地獲得短期大利益。

這是好方法去驅動人，但也不是好方法去成長。像韓非子所說，每個人都有貪圖的事物，善用觀察每個人貪圖的隱性價值，才是好方法去成長。

7.2 AI生產力，提高跨領域宏觀交易收益 AI Productivity to Enhance Cross-Domain Macro Trading Returns

當AI可以落地，輕量化的戰略執行。內部職能：①託管（Custody）、②合規（Compliance）、③報價（Pricing）、④估值（Valuation）、⑤審計（Audit）。對外溝通：IR、季度報告（Quarterly reports)、策略更新（Strategy updates)、投資人路演（Investor roadshows)。

第一，把內部職能轉化為宏觀收益流因子。

關鍵字 Core Principles	內部職能 Internal Communication	外部溝通對象 External Communication	輸入 Inputs	成果 Outputs	宏觀收益 Macro Returns	遠離危機 Crisis Avoidance	公式 Formula
安全性 Security	託管 Custody	託管銀行 Custodian Bank	資產清單 Asset Inventory	資產餘額 Assets Under Custody, AUC	流動性 Liquidity	對手風險、操作風險 Counterparty Risk, Operational Risk	Σ(持倉數量 * 收盤價格) + 現金餘額 Σ(Position Quantity * Closing Price) + Cash Balance
合規性 Security	合規 Compliance	合規顧問事務所 Compliance Advisory Firm	資產風險權重 Asset Risk Weight	杠杆比率 Leverage Ratio	杠杆 Leverage	監管風險、爆倉風險 Regulatory Risk, Margin Call Risk	(資產餘額 + 融資) / 自有資金 (Asset Balance + Financing) / Own Capital
準確性 Accuracy	報價 Pricing	數據供應商 Data Provider	每日收盤價 Daily Closing Price	收盤價格 Closing Price	匯率溢價 FX Premium	報價偏差、流動性風險 Quotation Deviation, Liquidity Risk	當日交易所收盤報價 Exchange Closing Price of the Day
透明性 Transparency	估值 Valuation	估值服務商 Valuation Service Provider	資產公允價值 Asset Fair Value	基金淨值 Net Asset Value, NAV	透明度 Transparency	估值錯誤、績效失真 Valuation Errors, Performance Distortion	(資產總值 - 負債) / 基金份額 Exchange Closing Price of the Day
監管性 Regulatory	審計 Audit	審計事務所 Audit Firm	歷史交易 Historical Transaction Records	年度審計報告中的淨資產 Audited Net Assets	信任 Trust	財務造假 Financial Fraud	審計確認的資產總值 - 審計確認的負債 Audited Total Asset Value - Audited Liabilities

第二，成本收益比率。錢要花得有價值。

淨利潤提升 = 降低工作時間（AI替代人工） + 增強盈利因子。
Net Profit = (Trading Gains + FX Premium + Trust Premium) − (AI Costs + Operating Costs)

把成果化为实际收益。不要只停留在展示数字。

淨利潤 = (交易收益+彙率溢價+信任溢價) - (AI成本+運營成本)。
Net Profit Increase = Reduced Working Hours (AI replaces manual labor) + Enhanced Profit Factors

內部職能 Internal Communication	雲服務成本端 Cloud Service Cost	改變公式中的變數 Change Variables in Formula	最終結果 Final Result
託管 Custody	資產清單管理 Asset Inventory Management	持倉數量↑、現金餘額↑ Position Quantity ↑, Cash Balance ↑	AUC更高 Higher AUC
合規 Compliance	AI合規引擎 AI Compliance Engine	融資額度↑、自有資金利用率↑ Financing Limit ↑, Own Capital Utilization ↑	杠杆比率優化 Optimized Leverage Ratio
報價 Pricing	數據供應商費用 Data Provider Fees	收盤價精度↑ Closing Price Accuracy ↑	報價偏差更低 Lower Quotation Deviation
估值 Valuation	AI估值模型 AI Valuation Model	資產總值↑、負債↓ Total Asset Value ↑, Liabilities ↓	NAV更高 Higher NAV
審計 Audit	AI審計工具 AI Audit Tools	融資成本↓ Financing Cost ↓	投資人信心更高 Greater Investor Confidence

About Me

My name is Kenny Chan. I work as a macro trader and structured broker, and I also use QuantAI tools as part of the AWS Community Builder.

I focus on global markets, manage profits, and aim to build a strong career in trading. My main skills include macro trading, structured brokerage, AI-based solutions, and risk management.

中港台「新经济，新金融」路演有感 Reflections on the “New Economy, New Finance” Roadshow

Kenny Chan — Tue, 16 Sep 2025 01:40:05 +0000

2025年8月29日，最后一站台湾，路演有感。

核心观点

1️⃣港台文化差异：新旧时代的轮舞曲；错失了优质力时机；高CP值的国民幸福感；高身份认同感；
2️⃣金融FSI行业篇：注重「整体收益」的高收益比；「复合型人才」是技术行业的大趋势；「领域+技术」双窝轮引擎；
3️⃣量化对沖基金篇：「罗马方阵vs日本浪人」文化；「逆人性」才是赚钱关键；创意择优；人机协同；安全可控；
4️⃣我学到了：人生聚焦，少即是多；以「深度」带出「广度」；但「细节」不等于「深度」；人生有无限种可能性，而不只有一条所谓「最好的路」；
5️⃣亚马逊领导力+桥水原则，基於现实的正确思考工具：每人眼中都有一个现实，因此每个「现实」各有不同；现实是指「务实」，而错误地理解现实，走向失败的「极端思考」。

聰明交易，迅速決策：以 AWS AI 洞察市場、發掘 Alpha 機會

本議程邀請經驗豐富的 Macro Trader 與 Crypto Trader 進行分享，親自分享如何實際採用 Amazon Bedrock、Amazon Q Developer，甚至 Kiro 等現代化工具，幫助交易者進行資料蒐集及決策制定，有效將市場噪音數據減少 80%，並提升超過 75% 的決策品質；同時也會由 AWS 專家分享，如何透過 AWS 工作有效發掘因子。

Smarter Trades, Faster Decisions: Unlocking Alpha with AI-Powered Market Insights on AWS

台湾有感

1️⃣新旧时代的轮舞曲；台南拥有传统的工业生产业，而台北拥有高科技创新业。而背后是年轻人对未来的决择，「往北/往南」都是一条好出路。
2️⃣错失了优质力时机；台币汇率曾贬值，虽然以一拆四贬值，以出口来解台湾的短期问题，但副作用也相当明显：台湾人不用进口货，不出国旅游，更没有动力去以高质量的目标经营事业，导致目前停留在「以量换价」的代工/中间品流程，因此错失了发展自己的台湾独立品牌的机会，更错失了发展「优质生产力」的最好时机。
3️⃣国民幸福感是可选择自己喜欢的生活风格；在台湾的街头有很多国民级娱乐，当中甚至有自发性的小团队，比如提「礼宾枪操/街舞/地雷系/亚系」等。台湾年轻人，随心选择自己喜欢的生活风格，而不用考虑别人的评价，比如「薪酬/晋升/专业性」的评价。因此，台湾是一个高CP的宜居城市。
4️⃣台湾有高身份认同感；反观香港人的身份认同感极低，甚至没有「国家/民族性」的概念，因此对自身的定位相当蒙糊，恰好反映在：虽有「国际视野」但仍「传统观念」办事，虽是「英语+普通话」教育但双语能力极不流利。而台湾人的身份认同感极高，心态上就显得从容，就算语言不好，往南也是一条好出路，因此人生不蒙糊也不焦虑，总有下一步可以选择。

精彩语录

金融FSI行业篇

1️⃣注重「整体收益」，而非「局部优化」；在技术行业，技术人有一个通病，很注重优化某一部份行业代码，预期提升0.01%准确度。但事实上，这种局部优化是极低收益。因此，技术人应该用大局思维去思考「整体收益」，或许优化某一点不起眼的技术流程有更高收益，比优化那0.01%准确度行业代码。
2️⃣「复合型人才」是技术行业的大趋势；COBOL是金融行业最早使用的编程语言，虽然是积累了四十多年的稳定业务经验，但随着愈来愈欠缺开发COBOL的技术人，因爲COBOL的冗长又难以维护；而Python/Java同时兼具「开发+数据分析+统计」等多功能属性，即孕育了同时懂「行业+业务+技术」的复合型人才，从而能选择自己喜欢的「行业/工作内容」，这是技术人的未来大趋势。
3️⃣「领域+技术」双窝轮引擎；纯粹的做技术，或者是纯粹的商业，其实都是盲人摸象，看不到行业的完整面貌。因此，行业专家都有双窝轮引擎，懂行业懂技术，才能实现「技术可以落地」，以各种「风险+短板+可行性」等维度，来做出行业价值。
4️⃣在新金融时代下，台湾金融法规显得相保守主义；台湾金融行业需要派人去观察机房，而香港的DataCentre机房，设有各种风险等级的区域，已是合符国际法规标准，所以就不用派人去观察机房。这背后显示了台湾法规更保守主义，如果能放宽限制，有助推动金融行业请向高质科技化。

量化对沖基金篇

5️⃣罗马方阵vs日本浪人，两个对冲基金文化；前者是把所有人整合到同一个portfolio更注重整体风险和透明度，互相分享知识+经验；后者是几个人用一个portfolio，注重个人独立的技艺，更具进攻性和威胁性，资本方可选择最出色的portfolio。
6️⃣「逆人性」才是赚钱关键，自己帐户vs别人帐户；如果是同一个交易员用同一个策略，惟一分别是「自己帐户vs公司帐户」，结果是自己帐户会亏钱。因爲人性令交易员产生「规避风险」心理，结果就是影响表现亏钱。
7️⃣创意择优，不在意用什么工具；无论是用传统的统计方法，还是最新的另类回测/因子发掘，还是最前沿的深度学习，只要最终效果是良好的，其实用什么工具都行。
8️⃣人机协同，量化虽好但不可缺失人类直觉；虽然用数学量化是确保了稳定性，但在关键时刻，仍然需要人类的直觉。
9️⃣云端只是「弹性成本」而非「安全可控」；对于金融商业客户，AWS的优势只是「弹性成本」，但现实中很多商业考虑，不单单是成本问题，比如有更重要的「可控」问题等，因此AWS不是首选方案。
量化对沖基金篇只是「行业洞察」，而不是最终答案。因此，只需选择对自己最好的独立观点，然后在现实中施展出来。

我学了什么？

1️⃣人生聚焦，少即是多；我聚焦在「宏观交易者」用「深度学习」在「主经纪商」。从而，在职场上的定位更加鲜明清晰，精准地切入了金融领域的「复合型人才」。
2️⃣以「深度」带出「广度」；当把「宏观」做的深入时，其实别人自然会联想到其他「交易领域」，即是以「深度」带出「广度」，而不用刻意强调广度「自己懂很多」。
3️⃣「细节」不等于「深度」；深度是有高价值的操作，比如是「带领往前瞻性目标/启迪伙伴提升层次」等。而非从细节入手，只是重覆性的低价值的操作，即俗称的「只懂玩细节，但连爲什么都解释不出来」。
4️⃣人生有无限种可能性，而不只有一条所谓「最好的路」；人生的目标不应该是「数字」一味地赚钱，而是把目标放在做觉得「有意义」的事情。幸运的是，我觉得做「交易+技术+创投+艺术」是有意义的事，并将其变成事业工作。

亚马逊领导力+桥水原则，基於现实的正确思考工具

①每人眼中都有一个现实，因此每个“现实”各有不同。

分析师喜欢沉默决策，把时间放在“沉淀/准备”上。强硬派讲话大声，把时间放在“结果/产出”上。外向派喜欢交流，把时间放在“人际/关系”上。

如何看不讲话？－－分析师沉默时，但外向派就在耳边吵“怎么不讲话？！”突然，强硬派大声主导新话题。
如何看讲话？－－外向派高谈阔论，但分析师觉得没重点，强硬派觉得没价值。

②现实是指“务实”，即向现实发动时间钳形攻击。

纯粹的“分析师/强硬派/外向派”都不现实。而现实应指务实，可以向现实发动“时间钳形攻击”，才能有好结果。
像索罗斯在二战期间的生存哲学，世界“变幻动荡”是真理，而“大稳定”才是诡异。因此，以动制动，人可以弹性地改变“分析师/强硬派/外向派”含量，从而发动“时间钳形攻击”有好结果。
但弹性地改变自己，需要强大的内心和正确的思考工具，否则只会在每次“改变自我”时做出错误决策。幸运地，“桥水原则”教导如何正确地思考，如何基於现实改变自己。

③错误地理解现实，走向失败的极端思考。

先讲“收费”是现实，但吹嘘成“1000w营业额”是不务实。“写代码”是现实，但想在代码世界找出“行业解法”是不务实。“认识大佬”是现实，但想靠人际关系学习“真工艺”是不务实。
做人既要现实，也要务实。可用“桥水原则”，把自己放低一点，把现实放高一点。即想象世界是一台“复杂和经常出错”的机械，如何用“现实世界”实现目标才是正确思考。

④朱熔基，“实务派”银行家，最佳实践“桥水原则”正确地思考。

朱熔基视中国宏观经济是一台“复杂和经常出错”的机械，针对中国90年代末的“政治环境丶宏观经济丶国际趋势”等因素，大刀阔斧地进行国企改革。
既要现实，也要务实。朱熔基同时结合了社会主义的“指令经济”做推力，和资本主义的“市场经济”做拉力，两条腿走路+摸着石头走路，发展出“中国特色社会主义的资本市场”。
朱熔基身上有各种鲜明又对立的形象：他既专横又谦虚，既聆听思考又语出惊人。他带领正确地“改变自我”的中国特色社会主义，也避免了“走向极端”的西方休克疗法。
既要现实，也要务实。是我在中港台之间看到的最明显差贯。

Postscript

Jacky Wu：非常高兴能在AWS台北的2025 AWS新金融丶新经济会议上与Danny and Kenny一同演讲！我们的议程「聪明交易，迅速决策：以AWS AI洞察市场丶发掘Alpha机会」深入探讨了现代工具如何改变交易领域。

About Me

Kenny Chan, Macro Trader, AWS Community Builder (Hong Kong), specialty of Fintech & Machine Learning

Macro Investments: Accelerating Market Data Decisioning with Bedrock & Q Developer [Tech for Trading series]

Kenny Chan — Mon, 28 Jul 2025 08:30:54 +0000

AWS Community Builder at AWS Summit Hong Kong 2025

In March 2024, I joined the AWS Community Builder in Machine Learning. Also, I am a macro trader, having worked in the “Sell-Side Macro Trading Desk Execution” and “Prime Brokerage” for nine years.

In August 2025, I became a speaker at the Dev Lounge at AWS Summit Hong Kong. I shared how to use the AWS Cloud AI service Bedrock to research macro market trends and suggest asset allocation portfolios.

My reflection at the AWS Summit Hong Kong 2025:

Marco Investments: Accelerating Market Data Decisioning with Bedrock on AWS [lv200,FSI,AI]

AWS accelerates the flow of global market data, but thousands of market data are volatile and uncertain.
Learn how Frontier Macro Trader processes market data rapidly and boosts trading decision profitability using Amazon Bedrock.
Explore Nova to underpin trading decision engines, build a knowledge base for consistent market insight, and use guardrails for high-quality market data.
Improve trading decision quality by over 75% and reduce market data errors by 80%.

Today, we are going to discover how to "Accelerate Market Data Decisioning with Bedrock & Q Developer on AWS".

I am Kenny Chan, a Macro Trader and AWS Community Builder.
Also, I specialize in GenAI financial infrastructure, tactical portfolio management, and macro trading desk execution.

What would happen to your home, if 99% of your bank savings disappeared?

The reality market is dramatically uncertain.

Fear: loses more than others; Greed: gains less than others.

People can't control their emotions.

In this session, you will find how to use Amazon Bedrock and Amazon Q to enhance your market data decisions.
Next 10 minutes, we will go through these 3 steps.

By the end of this session, you will find
① how to improve the quality of trading decision by over seventy-five %
② and also reduce noisy market data by eighty %.
So that, this session can help you transform your business thinking.

Someone may question: Why is Market Data Decisioning important to us?
Yes, because in today's uncertain world, having the right tools and even the right infrastructure can make all the business difference.
The good news is that AWS provides a advanced infrastructure to support market data decisions.

We first focus on step 1 Bedrock Decisioning.
And we will go through how Amazon Bedrock, OpenSearch, and S3 bucket as foundational elements in decision-making.

I believe in "Quality In, Quality Out".
So, I collect lots of Investment Banking Insights and My Historical Trading Journal.

Go to the OpenSearch tab, click the create button, pick the knowledge base with Vector Stores.

Fill in the Knowledge Base name.

Data source name and S3 URL

Meanwhile, choose the both Embedding model and Amazon OpenSearch Serverless.
Here we simply choose Titan Text Embedding.

After that, Pick the knowledge base we just-created.

Click the "Add documents" button to

Upload documents.
Here we just upload Investment Banking Insights and Ours Historical Trading Journal.

Next, choose the Bedrock agent tab, fill in the Bedrock agent name, and

Pick the bedrock agent model.
Here we simply choose Titan Text G1 Premier.

Define the prompt instruction for the bedrock agent.

Assign bedrock agent to knowledge base.

And done. Now, you can ask bedrock agent for market data decisions.

You can even ask bedrock agent for a marco portfolio.

Now, Amazon Q integrates with the AI workflow.
So that, if you forget any steps, Q allows you to quickly find answers.

Here are the examples for global market analysis

and muti-assets allocation.
For me, Amazon Bedrock is so helpful for my trading desk.

When you complete the usage, don't forget to delete the bedrock agent

delete the knowledge base

and delete the OpenSearch Serverless

We next focus on step 3 Q Developer Portfolio P&L

We will go through how Amazon Q Developer and Backtrade, which is a Python trading framework, can be used to backtesting Portfolio P&L.

Here is backtesting P&L. The most amazing thing is we did it in zero-code.

which means that, we just ask the Amazon Q developer and everything is done.

Also, here is the global-market correlation.

The same. we just ask the Amazon Q developer, and done.

Beautiful! You can find my GitHub to get the whole Amazon Q developer source code.

The monthly cost is 400 US dollar. It is a competitive price for both individual traders and Institutional investors.

Finally, we go back to Step 2, Macro Pure-Alpha Architecture.

Because everything in the world is uncertain, so the most important is your business thinking, which means how to build your dynamic architecture.

Here is the golden rule I found in my 10 years of trading experience.
No need always true, but need to know when will fail.

Thank you for joining me today.
① I encourage you to connect with my LinkedIn for further discussions,
② and to stay updated in financial services industry on AWS.

Postscript

I have graduated from the AWS NewVoices 2025 speaking program! LinkedIn

I am a Macro Trader and have never used #AWS Q CLI. Still, I created a backtrader python to #DeltaHedge #Indexing using only two prompts at the AWS Hong Kong for Capital Markets on 10 June 2025. LinkedIn

What an incredible day at our AWS Hedge Fund Innovation Day with Factor Modeling! 🚀 LinkedIn

Why did I choose AWS to grow my trading journey?

We're different. Only a few traders are profitable, so we need to be ① innovative, ② performance-driven, and ③ different from the majority.
And AWS leadership culture, AWS people, and AWS services all have these factors.

AWS x Macro Trader

Meaning	AWS	Trader
①Sun	AWS Leadership Culture	Pure Alpha
②Life	AWS People	Innovation & Brave
③Soil	AWS Services	Absolute Risk Control

This spirit of innovation is something I have never seen elsewhere. Thus, I chose AWS to enhance my trading journey.

Every successful trader is different, so AWS Bedrock is fragmented, allowing for the creation of an infinite number of workflows.

The following is the latest “macro market trends” AWS Bedrock sell-side research workflow.

Pure alpha x Absolute risk control

Commander ➡️ Asset allocation manager
Risk defense engineer ➡️ Risk manager
Intelligence reconnaissance officer ➡️ Investment analyst
Tactical assault captain ➡️ Trader

Macro hedge traders have three key alpha edges: ① Absolute net value, ② Macro timing, ③ Odds rotation.

Therefore, Bedrock must perfectly align with macro hedge trading to achieve outstanding performance results.

Reference: aws-samples / sample-tech-for-trading / investment_advisor_by_agentic_ai

Excellent Writing Communication: Commander Intent, Amazon Narrative, and Trading Strategy Notes

I publicly stole the “Amazon Six-Page Narrative😱” to reinforce my trading journey.

What is Commander Intent?

Commander’s Intent is a clear and concise expression of the ①purpose of the operation and the ②desired end state. It explains why the mission is being conducted and ③what success looks like, ④without prescribing how to achieve it in detail.

	Commander Intent
Purpose	1️⃣Our team needs to take control of the bridge so our friends can safely cross and stop the enemy from getting away. 2️⃣If we control the bridge, our side can move forward and win the battle.
Key Tasks	1️⃣Get to the bridge quickly 2️⃣Keep the enemy away 3️⃣Make sure our team can cross safely
End State	1️⃣The bridge is safe and clear 2️⃣Our team is moving forward 3️⃣The enemy cannot escape

How can ⭐Amazon culture⭐ be leveraged to reinforce trading desk performance?

I create strategic value by achieving goals through others. Therefore, I give others the greatest assistance, allowing them to execute tasks freely and independently, so that everyone wins.

Take control of the bridge so our team can move forward.
Even if the bridge is broken. Our team can find another way to cross, like using boats or finding another path.
This still helps complete the mission and reach our goal.

	My Trading Notes on April 7, 2025
Purpose	We want to *protect our money* by changing how we invest. On April 7, 2025, we sold our stocks and started using safer things like strong currencies and smart selling.
Key Tasks	1️⃣Use *short-term bonds* and long-term bonds. 2️⃣Buy *gold, copper, aluminum, oil, and U.S. dollars. 3️⃣Sell some things* without owning them first (called short selling).
End State	1️⃣The bond market is *tighter, so we moved to safer money tools on April 8. 2️⃣Stock prices are jumping up and down a lot. Some political talk made our selling plan not work well. 3️⃣Things that used to move together now don’t, so our plan to balance risk didn’t work*, and we lost some money.

About Me

Kenny Chan, Macro Trader, AWS Community Builder (Hong Kong), specialty of Fintech & Machine Learning

为什么我选择AWS来提升我的交易生涯？（中文版摘要）

我们不一样。只有少数交易者可获利，所以我们必需①创新、②绩效导向、和③大众不一样。
而AWS领导力文化、AWS人、AWS服务，都拥有以上因素。

AWS x Macro Trader

含义	AWS	交易员
①太阳	AWS领导文化	纯阿尔法收益
②生命	AWS人	创新与勇敢
③泥土	AWS服务	绝对风险控制

这种勇敢创新的精神，我未曾在其他地方见过。因此，我选择AWS来提升我的交易生涯。

在2024年3月，我加入AWS社区开发者（深度学习）。另外，我是一名宏观交易者，在“卖方宏观交易桌执行组”和“主经纪商”工作九年。
在2025年5月8月，我成为AWS Summit Hong Kong的Dev Lounge演讲者。分享用AWS云AI服务Bedrock，研究宏观市场趋势、建议资产配置组合。

以下2篇文章，是我的启发在AWS Summit Hong Kong 2025。

更新2025-07-28：

每个成功的交易者都不一样，因此AWS Bedrock是零碎，才可组合成无限个工作流。
以下是最新的"宏观市场趋势"AWS Bedrock卖方投研工作流。

纯阿尔法收益 x 绝对风险可控

指挥官➡️资产配置经理
风险防御工兵➡️风险管理师
情报侦察参谋➡️投资分析师
战术突击队长➡️交易员

宏观对冲交易者有三大阿益法收益，即是三大交易优势。 ①绝对净值 ②宏观择时 ③赔率轮动
因此，Bedrock要完美地契合宏观对冲交易，才能达到优秀的绩效结果。

优秀写作沟通：指挥官意图和亚马逊叙事和交易策略笔记

我公然窃取了"亚马逊六页纸叙事"来强化交易生涯。

如何利用⭐亚马逊文化⭐，强化交易桌绩效？

我创造战略价值，通过别人来实现目标。因此，给了别人最大的帮肋，让别人可自由地执行任务，最终大家都是嬴家。

	交易笔记
目的	维持2025-04-07(一)*积极防守战略*:清空全球权益配置，转移到在"货币属性资产+对冲沽空"。
关键任务	1️⃣DR0007>短国债>长票息>短价差 2️⃣黄金商品>铜铝>石油>美元指数 3️⃣*Naked short* selling
最终状态	1️⃣债市的流动性被压缩，在2025-04-08转向了“货币属性”资产。底仓这两天出现－2%的回撤。 2️⃣权益资产波动率*超预期*。川普的政治口术，令裸空部位的收益不够大。加上SP500两天-10%>+10%，隐含波动率伴隋不确定性上升，多空难做。 3️⃣宏观架构的资产相关性受弱化，资产对冲失效，所以出现净值回撤。

宏观投资： AWS Bedrock 加速市场数据决策 [lv200,FSI,AI]

AWS加速全球市场数据流动，但千亿笔市场数据是易变和模糊。
学习Frontier Macro Trader如何快速处理市场数据，用Amazon Bedrock提高交易决策盈利。
探索Nova支撑交易决策引擎; knowledge base建立统一的宏观市场观点; guardrails过滤有害的异常值市场数据;
提高交易决策质量超75%，和减少80%的市场数据错误。

Trader's Survival Race [AWS Q CLI Gaming]

Kenny Chan — Tue, 17 Jun 2025 03:24:00 +0000

🎂 Trader's Survival Race

① The best moneymakers tend to go against human nature
② World Race to Challenge Profitability
Website: https://amazon-q-macro-investment-002.s3.us-east-1.amazonaws.com/index.html

Game Description

You are a professional trader who has gained infinite wealth through high-frequency trading. However, professional trading is a survival game; only the top 20% of professional traders have positive returns. Therefore, the most profitable traders tend to go against human nature.
In the world race to challenge profitability, you will encounter the following two questions:
1️⃣Can you become a great trader through post-training?
2️⃣Is it possible to make a lot of money by being able to predict future prices?

🎀 Six Common Characteristics of Profitable Traders

1️⃣ Recognize Three Core Profit Advantages
2️⃣ Insist on Using the Trading Style You're Best At
3️⃣ Protect Capital
4️⃣ Strict Stop Losses
5️⃣ Be brave enough to place big bets at high odds
6️⃣ Know You Were Wrong, and Don't Make Wrong Again
Conclusion: A good mindset is a determining factor in whether or not you can become a great trader; therefore, with the right training, anyone can be a profitable trader. Anyone can become a profitable trader.
However, a bad mindset can easily destroy a good trader.

🎁“Future price” is not important; the “present moment” is the most important

①Soros's Quantum Tiger Macro Hedge Fund, with 30 years of trading accuracy of about 42.56%, in other words, with a coin flip way to buy and sell, is more accurate than Soros.
②However, Soros' Quantum Tiger Macro Hedge Fund has an annualized return of 22%, which means it doubles every five years. The meaning is that, in addition to accuracy, trading skills such as “high odds + macro hedging + asset allocation + long/short leverage + narrative analysis” are the core advantage of making big money.
③Therefore, focusing on “factors that can be controlled at the moment” is the core profit advantage; and “~~predicting future prices~~” is not the most important.

📈 My core profit advantages are

1️⃣ Macro Hedge Architecture: a “all-weather risk parity” strategy through multi-asset diversification of risk
2️⃣ The Art of Slow: “Fixed Income +” Make Money Slowly, Focusing on Risk Boundaries
3️⃣ Idea Meritocracy: focusing on the quality of the trade concepts only and not caring about the person they come from

📝How to play

1️⃣Watch the K-chart of the product
2️⃣Select the product, fill in the quantity, click buy/sell
3️⃣Total returns will be shown automatically

🥳Surprise points of using Q CLI

1️⃣Q CLI automatically generates notes on financial terms such as “Realized P&L / Unrealized P&L / Annual return”.
2️⃣Q CLI draws beautiful front-end HTML. Because of the visual model, and step-by-step instructions, Q CLI works quite well.
3️⃣Provide appropriate technical advice. For example, Q CLI suggests using Tradingview's light-chart front-end framework library.

😭 Frustration points of using Q CLI

1️⃣Can't understand financial algorithms such as “average buy price / net position held” etc.
Solution: provide algorithmic formulas to Q CLI.
2️⃣ People don't understand certain professional terms.
Solution: Learn the domain yourself and then use Q CLI to improve industry productivity. Because Q CLI knows a lot of domain knowledge, but users often do not know how to give accurate commands.
3️⃣ Financial treasurer only uses Excel to work, there is no need to use Q CLI and other AI tools.
Solution: Q CLI is a human assistant that helps human industry experts as an “assistant”, such as “collecting information/organizing data/checking reports”. It does not disturb the human industry experts while providing assistant services to enhance productivity.

🎂交易者的生还竞赛

①最会赚钱的人，往往都是逆人性
②挑戰盈利能力的世界競賽
立刻试玩：https://amazon-q-macro-investment-002.s3.us-east-1.amazonaws.com/index.html

游戏简介

你是一名职业交易者，通过高频交易积累无限财富。然而，职业交易者是生存游戏，最终只有顶尖20%职业交易者拥有正收益。所以，最会赚钱的交易者，往往都是逆人性的。
在挑戰盈利能力的世界競賽里，你将会遇到以下两个问题：
1️⃣可否通过后天训练，成爲出色的交易者？
2️⃣能预测未来价格是否能赚大钱？

🎀盈利交易者的六项共同性格

1️⃣认清三项核心盈利优势
2️⃣坚持使用最擅长的交易风格
3️⃣保护本金
4️⃣严格止损
5️⃣勇于下重注，在高赔率的时机
6️⃣知道自己错了，不再犯错
结论：好心态，能决定能否成爲出色的交易者；因此，通过后天训练，任何人都能成爲盈利交易者。
但是，坏心态可轻易摧毁一个出色交易者。

🎁未来价格不重要；「当刻」才是最重要

①索罗斯的量子老虎宏观对冲基金，30年交易的准确度约42.56%，换言之，用掷硬币方式买入卖出，都比索罗斯准。
②但是，索罗斯的量子老虎宏观对沖基金年化收益率有22%，即是每五年升幅一倍。意味着，除了准确度之外，「高赔率+宏观对冲+资产配置+多空杠杆+敍事分析」等交易技巧，才是赚大钱的核心优势。
③因此，关注「当刻可控制的因素」才是核心盈利优势；而「~~预测未来价格~~」不是最重要的。

📈我的核心盈利优势是

1️⃣宏观对冲架构：通过多资产分散风险，实现全天候「风险平价」策略
2️⃣慢的艺术：「固收+」慢慢地赚钱，注重风险边界
3️⃣创意择优：只关注交易概念的质量，而不在乎出自谁

📝如何玩

1️⃣观察产品的K线图
2️⃣选择产品，填写数量，按买入/卖出
3️⃣总收益将自动显示

🥳使用Q CLI的惊喜点

1️⃣Q CLI自动生成金融术语的注解，如「Realized P&L / Unrealized P&L / Annual return」等。
2️⃣Q CLI能绘画优美的前端HTML。因拥有视觉大模型，逐步指令Q CLI的效果相当不错。
3️⃣提供合适的技术建议。比如Q CLI建议使用tradingview的light-chart前端框架库。

😭使用Q CLI的挫败点

1️⃣不能理解金融算法，比如「平均买入价 / 持仓净值」等。
解决方法：提供算法公式给Q CLI。
2️⃣大家都不懂某个专业术语。
解决方法：自己先学习这个领域知识，然后使用Q CLI提升行业生产力。因爲Q CLI是懂很多领域知识，但使用者往往不懂发出精准的指令。
3️⃣金融财务员只用Excel工作，没需要用到Q CLI等AI工具。
解决方法：Q CLI是人类助手，以「助理」身份帮助人类行业专家，如「收集资料/整理数据/检查报表」等。既不打扰人类行业专家，同时提供助理服务提升生产力。

About Me

Kenny Chan, Macro Trader, AWS Community Builder (Hong Kong), specialty of Fintech & Machine Learning

Reference

Prompts

1️⃣AWS Titan text-to-image

Game Name: Trader Survival 
Game Genre: Simulation, Strategy 
Target Audience: 18-40 years old financial trading enthusiasts.
Cover Design Concept: 
Color: Dark blue and grey background with yellow and green accents.
Graphics: Digital chart trading floor in the background, trader operating a computer in the center surrounded by financial icons.
Composition: trader in the center, symmetrical layout, three-dimensional metal font game name on top.

2️⃣AWS Q CLI Gaming

1. create index.html, use "https://unpkg.com/lightweight-charts@4.2.1/dist/lightweight-charts.standalone.production.js" to draw chart, and generate random time-series data to tradingview data feed
2. index.html, add random new hloc data so update the chart in every 1s
3. index.html, modify to 12 tradingview chart and contains all js function
4. index.html, add "https://cdn.tailwindcss.com" and change to use tailwindcss
5. index.html, lightweight charts:
[] handleScroll: false and handleScale: false
[] add chart title with product 1 to 12
6. index.html, change all ui to dark mode
7. index.html, add new container on the left side, which is statistics layout:
[] volatility table, show every products 10 snap volatility in percentage
8. index.html, add paper trading panel under the bottom, includes
[] product horizontal single-select listbox, qty, buy & sell button
[] position table and updated position P&L (avg bug price - product current price) in every 200ms
[] transaction table
9. index.html, modify html
[] modify Positions table & Transactions table in 2 cols
[] modify Transactions table, scroll table height 300px
10. index.html,
[] updated position P&L ((product current price - avg bug price) * Quantity)
[] add total P&L
11. index.html
[] move "total P&L" on the right of "Paper Trading" title
12. index.html
[] modify "total P&L" formula. "total P&L" also count on "Transactions Transactions".
13. index.html
[] always show "total P&L". default is 0.
14. index,html
[] add sell logic, show error if sell Quantity larger than Positions holding Quantity

AWS是多模态组织；以「企业，行业专家，创投家，开发者」视角敍述「AWS商业技术」

Kenny Chan — Sun, 11 May 2025 09:36:50 +0000

AWS是多模态组织；以「企业，行业专家，创投家，开发者」视角敍述「AWS商业技术」

📝结论

1️⃣ AWS是「硬件/轮子/宣发/创造」
在企业视角，AWS是「硬件」；提供云服务中游设备。
在行业专家视角，AWS是「轮子」；提供商业技术的基础元素。
在创投家视角，AWS是「宣发」；提供路演宣传渠道。
在开发者视角，AWS是「创造」；提供工具实现开发梦想。

角色视角	AWS属性	AWS用途
企业	硬件	云服务中游设备
行业专家	轮子	商业技术的基础元素
创投家	宣发	路演宣传渠道
开发者	创造	工具实现开发梦想

2️⃣ AWS体验差异之大，因用户的需求各不相同
AWS是多模态企业。虽然这样可满足不同的「硬件/轮子/宣发/创造」需求；但是，「多模态」也意味着蒙糊不稳定，因此体验差异极大。
因此，在AWS应追求「蒙糊的正确」，而非精准的答案。

3️⃣ AWS用户群有「批评者/跟随者/参与者/分享者/构建者/创造者」
批评者：恶意打击AWS的缺点不足；不应理会这种恶意
跟随者：跟随舆论来判断AWS好坏；应主动推广AWS的「意义/理由/动机」
参与者：出席AWS的活动讲座工作坊；应多派发「纪念品/优惠券/考证教材」
分享者：经常转发AWS的内容；应给予「推广大使/营运大使」头衔
构建者：用AWS技术来开发应用工具；应多给予「硬件/轮子」支持，并且多联络他们
创造者：用AWS技术来创造商业生意；应给予「轮子/宣发」支持，并且避免打扰到他们

用户定位	定义	行动
🙅批评者	恶意打击	不应理会
🙍跟随者	跟随舆论	推广「意义/理由/动机」
🙆参与者	出席活动讲座工作坊	派发「纪念品/优惠券/考证教材」
🙋分享者	转发内容	给予「推广大使/营运大使」头衔
🤷‍♀️构建者	开发应用工具	给予「硬件/轮子」支持，并且多联络他们
🙇‍♀️创造者	创造商业生意	给予「轮子/宣发」支持，并且避免打扰到他们

4️⃣ 开发者/构建者聚焦应用价值；而创造者聚焦创造价值
开发者/构建者聚焦技术；把技术视爲一门「工艺」，来实现应用价值。因此，他们既懂代码技术看论文，又热爱交流想法心得。所以，应多给予「硬件/轮子」支持，并且多联络他们。
创造者聚焦价值，把技术视爲一门「工具」，来实现目标价值。因此，他们既懂商业又懂科技应用，但未必懂代码技术看论文，也未必热爱分享创业ideas/自己的正职工作。所以，应给予「轮子/宣发」支持，并且避免打扰到他们。
但是，也有少数人同时有「开发者+创造者」特质；即热爱分享创业ideas的「商业技术极客」。

角色视角	聚焦点	技术是什么	定义	行动
开发者/构建者	技术	工艺	①懂代码技术看论文，②交流想法心得	①应多给予「硬件/轮子」支持，②并且多联络他们
创造者	价值	工具	①既懂商业又懂科技应用，②但未必懂代码技术看论文，也未必热爱分享创业ideas/自己的正职工作	①应给予「轮子/宣发」支持，②并且避免打扰到他们
商业技术极客	商业技术	人的价值	①懂代码技术看论文，②热爱分享创业ideas	①应多给予「硬件/轮子/宣发」支持，②并且多联络他们

5️⃣ Amazon的「聆听->学习->帮助->领导」原则
无论AWS是「硬件/轮子/宣发/创造」，还是用户群有「批评者/跟随者/参与者/分享者/构建者/创造者」；通过「聆听->学习->帮助->领导」原则，可追求蒙糊的正确。

6️⃣ AWS有八万四千法门，只需寻找属于自己的法门
AWS聚焦「人的价值」，通过超过200个服务作为基础轮子，帮助所有人实现自我价值和目标。
在AWS，最重要是「实践出自己的答案」，而非通过别人爲自己提供答案。
因此，AWS有很细碎的服务，也是蒙糊的多模态企业，来满足不同人的不同「商业技术」需求。

🍿案例

1️⃣商业与技术之间的鸿沟，可通过「细化个性标签/产品力/故事力/行业前沿技术」弥补

支点	身份	目标听众	定义
细化个性标签	行业专家	企业家，潜在客户	深化聚焦三个前沿专精技术，通过标签化可减少技术专家「不太技术，不太商业」的蒙糊视感
产品力	产品经理	组织内部（如合伙人，高级经理）	通过产品UI来弥补「技术+商业」的鸿沟
故事力	技术销售	外部人员（如潜在客户，其他行业专家）	因外部人员不懂，所以可通过「Storytelling」讲故事来传播理念
前沿行业技术	创投技术家	投资人（有限合伙人，潜在领投资本）	用前沿行业技术爲切入点，打造商业差异化，实现募资需求

在「斜杠人生+乌卡时代」，高管必需学习以上的技能，才能获得优秀的商业技术思维，以应对急速进步变化的新时代。

2️⃣领导力，是指「三分钟」决策大事，「三天」处理重要小事

支点	表现	定义
领导力	愿景感受	三分钟决断未来十年大事，拥有「极端开放思维+承担失败的勇气」来打造应对多变的优秀企业
管理力	流程品质	把控好「高风险/不畅顺」的节点，通过「聆听->学习->帮助->领导」来打造良好的工作环境
执行力	工具方法论	形成一套「方法论/行业标准」，然后持续优化使用，提高生产力和品质

高管需有「抓大放小」的能力，同时兼备「领导力，管理力，执行力」，才能用三分钟决断大事，三天处理重要小事。

✅Good高管角色	❌Bad高管角色	😈Bad高管表现	😈Bad高管金句
行业专家	传销	行业领域「0触觉」	①别人都是AI硬科技，现在主题是Next GenAI ②您看一下这些热点议题
产品经理	监工	只能把控「时间表」	①大家负责好自己的事，下周给进度 ②您可以先填一下XX表格，报一下YY的ZZ活动
技术销售	中介	只能对接「人脉资源」	①给一下XX方案链接 ②您跟XX沟通一下
创投技术家	助理	「数字KPI」导向价值	①一千万融资，三十万活跃用户 ②稍晚点跟您同步，我现在在外面跟Y总聊

拥有高管格局的的「领导力，管理力，执行力」，那么一个优秀企业就拥有「行业专家+产品经理+技术销售+创投技术家」。否则，只是三流公司的「传销+监工+中介+助理」。

参考：高管领导的六个关键特质 - 金水杯奖

3️⃣多模态组织是「协作关系」，而非「上下层级」关系

4️⃣对抗性合作，是像碰碰车「共同前进」

5️⃣人的价值>商业价值，从「人的价值」找出最优答案

6️⃣社区=工会，是员工与企业之间的「桥樑」

About Me

Kenny Chan, AWS Community Builder (Hong Kong), specialty of Fintech & Machine Learning

有多少爱可以重来；AWS香港高峰会路演2025Q2感受

Kenny Chan — Sat, 10 May 2025 10:33:15 +0000

💼高管领导的六个关键特质

📝序章

2025年后疫情时代，是「斜杠人生+乌卡时代」爲叙事主线。前者是同时有更多职业角色，后者是环球的政治经济不确定性。

通过香港亚麻高峰会2025Q2，探索①如何快速提升个人价值，②如何规划好未来十年职涯，③如何选择好职涯中的商业技术定位

📢关键字：#向内聚焦 #钝感力 #故事力 #商业技术 #蒙糊正确 #绝对理性

1️⃣斜杠人生更要向内聚焦；爲职涯选择三个标签，再向内聚焦。因行业专家是「技术里不太技术，商业中不够商业」的蒙糊视感，所以向内聚焦三个标签。当客户通过特别聚焦的标签而找到行业专家时，然后自然把商业项目从这一聚焦标签扩展开来。

2️⃣钝感力是应对乌卡不确定时代的重要技艺；在人生建立心灵屏障来锻练钝感力。因职场上大家有不同的观点，钝感力可令你更聚焦在目标，通过「合作性竞争」来实现win-win共同进步的最优结果。

3️⃣故事力是高管领导传播愿景的手段；愿景是体现「人的价值」，反而和商业技术资本等没关系。因高管领导是引领企业的角色，不能只从「管理/执行」的观点上来制定工作标准流程。因此，故事力是传播「企业核心价值文化」的最高效沟通工具。

4️⃣商业技术实现愿景的是「工具/手段/资源」；所以商业技术应并存，而非割裂。因人是构成商业社会的最基本元素，由此「商业技术」是一个工具来发掘「人的价值+商业愿景」。把一个商业护城河做专精，同时也要发掘人的价值。

5️⃣蒙糊正确可使人勇往直前；过份精准只会难以承受突如风险。因在乌卡不确定时代，所以「弹性积木思维」才能应对突如风险。作为高管领导，更需要有「极端开放的思想+承担错误的勇气」。

6️⃣绝对理性是指「不情绪化」；高管需要有稳定理性的表现，才可以领导好「商业运作+突给其来的风险」。因企业社会是由人来构成，而KPI目标只是高管「绝对理性的手段」，但「人的价值」才是商业价值的最终愿景。

🔎案例

1️⃣ 高管如何领导好重大事情？
香港某金融公司有一个叫「金水杯」的最佳交易员奖。
因有位交易员杰出表现，所以领导Feedback Loop即时奖励，而手中刚好只有只水杯，就用来表扬这位交易员的杰出表现。从此，公司就有了「金水杯奖」。
作为高管，把握「时效性」很重要。不要忽略团队气氛这些看似小事，但其实这些是企业大事。
作为高管，果断地用三分钟决定千万额生意；但却用三天细腻地处理「内部沖突/撰写信件/演讲宣发」这些看似小事，但实际上这些是企业大事。

2️⃣ 如何规划好行业专家的职涯？
行业专家被认爲是「技术里不太技术，商业中不够商业」，主要是不够聚焦自己的护城河。
比如「金融+投资+创投」是三个高维度的标签；因此，更深入聚焦到「风险管理+宏观对冲+GenAI创投」细緻的个人标签，以「less is more少即是多」爲职涯方向。
「向内聚焦+多向内求」三个职涯标签；当客户通过特别聚焦的标签而找到行业专家时，然后自然把商业项目从这一聚焦标签扩展开来。

🕺行业专家的GenAI最优解，以金融行业爲例

📝序章

2025是GenAI井喷之年，但据调查显示，行业专家对GenAI感到困惑。一方面，GenAI的「可解释性」偏低，所以GenAI在高专业性行业表现不佳；另一方面，高专业性行业的「工作执行文化」，导致GenAI 未能应用到工作里。

通过香港亚麻高峰会2025Q2，探索①如何提高GenAI在高专业性行业的可解释性，②如何利用GenAI提升高专业性行业的价值产出，③如何让高专业性行业人员学习GenAI

📢关键字： #高专业性行业 #行业专家 #可解释性 #商业思维 #价值产出

1️⃣拥抱高专业性行业有独特的「工作执行文化」；金融风险管理是处理「突发意外+不确定性」。因此仍是人力主导的工作，所以GenAI的应用率仍很低。

2️⃣行业专家的视角是共同标准；不是单靠增加「人力/成本/时间/技术/资本/方案」就能解决行业痛点。金融风险管理有共同标准，如使用传统VaR算法来控制资产组合风险，因此额外增加资源不会有正面改善。

3️⃣在「突发意外+不确定性+共同标准」的高专业性行业， GenAI可作为人类助理来提供「大致正确的建议」；

4️⃣GenAI的商业价值，是帮助行业专家提升「个人价值」，而非取代人类。

5️⃣GenAI在高专业性行业，更似是汽车业发展。虽然车速快了，好象更危险了；但现代汽车比以前更安全，对人类生活也更美好了。

🔎案例

1️⃣ 如何利用GenAI提升高专业性行业的价值产出？
金融法规是文书场景，主要是用Word工作。而金融风险是试算表场景，主要是用Excel工作。而财务监管是财报场景，主要是用PDF工作。
因爲高专业性行业的独特执行文化，令GenAI难以直接取代人，同时难以把GenAI应用到工作场景；另外，高专业性行业也难以更换「办公软件/场景/流程」，通过额外增加资源。
因此，GenAI是人类助手，通过帮助行业专家减少重覆工作，或减少人爲的工作错误；从而提升个人价值。

2️⃣ 如何让高专业性行业人员学习GenAI？
高专业性行业，意味着行业专家的「每token是极高价值」；所以，行业专家缺乏诱因学习GenAI。
另外，GenAI取代了基础的人类工作，比如是基础全栈开发/数据开发，或基础金融算法/财务分析/风险管理；因此，对于想进入IT/金融行业的最低门槛，反而是先学好AI生产力工具。
但对于行业专家来说，人生护城河不是着重在「执行/管理」岗位，而是在「领导力」的行业感知上。
比如，香港某金融公司因只有99.99%时间有效模型，所以花钱招聘行业专家，才弥补那0.01%的风险；这就是人类行业专家的优势，而不是通过「GenAI/技术/算法」来建立优势。

参考： AWS多模态组织 - 案例2️⃣ 领导力，是指「三分钟」决策大事，「三天」处理重要小事

❤️有多少爱可以重来；香港亚麻高峰会路演2025Q2感受

That’s a wrap for AWS Summit Hong Kong 2025 as a NewVoices DevLounge Speaker! LinkedIn

Happy birthday on the occasion of AWS Summit HongKong 2025 DevLounge! LinkedIn

What an amazing day at the DevLounge Felicitation Dinner of AWS Summit Hong Kong 2025! Linkedin

I have graduated from the AWS NewVoices 2025 speaking program! LinkedIn

About Me

Kenny Chan, AWS Community Builder (Hong Kong), specialty of Fintech & Machine Learning

AWS Best Practice of DevOps Agile Delivery for the Financial Services Industry

Kenny Chan — Fri, 27 Dec 2024 07:17:35 +0000

The previous chapter 《AWS DevOps Agile Delivery of 16 Leadership Principles for the Financial Services Industry》shared how AWS DevOps pipelines can solve pain points in the financial services industry, and utilize the Amazon 16 Leadership Principles.

In this chapter, you will learn how to build an AWS DevOps pipeline:

AWS Services	Description
IAM	Identity and Access Management
EC2	Cloud-computing platform
Elastic IP address	Static IPv4 address designed for dynamic cloud computing
Route53	Cloud domain name system (DNS) service
CodeDeploy	Automate application deployments to Amazon EC2 instances
GitHub Actions	Easy to automate all your software workflows
Pricing Calculator	Create an estimate for the cost of your use

2.0 AWS DevOps Pipeline

2.1 Pre-requisites

2.1.1 Knowledge Pre-requirements

Create an EC2 server
Have a GitHub account and know basic Github Actions.
Know how to setup NGINX
Know basic AWS services, including EC2, CodeDepoly, IAM.

2.1.2 Project Requirements

First upload a simple static web project codedeploy.nginx.001 on Github, which includes:

Object	Location
index.html	./	Static Web Page
ic_alana_002_20241022_a.jpg	./icons	images on a static web page
appspec.yml	./	CodeDeploy code
application-stop.sh before-install.sh after-install.sh application-start.sh validate-service.sh	./scripts	CodeDeploy code
appspec.yml	./github/workflows	CodeDeploy code

Also, GitHub access tokens are needed to configure codeDeploy permissions.

Github -> Setting -> Developer Setting -> Tokens. Add a GitHub access token.

2.2 Creating IAM Roles

A good naming style is important because as the number of IAM roles grows, it can be confusing for developers.

AmazonSageMaker-ExecutionRole-20240805T101031
AmazonSagemakerCanvasBedrockRole-20240801T140683

{service}-{role}-{datetime}-{version}. AWS Bedrock and SageMaker auto-generated IAM naming style.

AWSCodeDeployService-EC2AccessCodeDeployRole-20241024T000000
AWSCodeDeployService-DepolyEC2Role-20241024T000000
AWSCodeDeployService-GitAssumeRoleWithAction-20241024T000000

This is the clear IAM naming style, so we will create three IAM roles for EC2, CodeDeploy, and GitHub Actions, respectively, following this official IAM naming style.

2.2.1 AWSCodeDeployService-EC2AccessCodeDeployRole-20241024T000000

Select EC2 on Use Case Tab。

AmazonEC2FullAccess
AmazonEC2RoleforAWSCodeDeploy
AmazonS3FullAccess
AmazonSSMManagedInstanceCore
AWSCodeDeployFullAccess
AWSCodeDeployRole

Add AmazonEC2, AmazonS3, and AWSCodeDeploy permissions.

2.2.2 AWSCodeDeployService-DepolyEC2Role-20241024T00000

Select CodeDeploy on Use Case Tab.

AWSCodeDeployFullAccess
AWSCodeDeployRole

Add AWSCodeDeploy permissions.

2.2.3 AWSCodeDeployService-GitAssumeRoleWithAction-20241024T000000

Select Access management -> Identity providers -> Add provider.

Used to listen to GitHub Actions.
Provider URL: token.actions.githubusercontent.com
Audience: sts.amazonaws.com
The GitHub Identity Provider then adds the AWSCodeDeployService-GitAssumeRoleWithAction-20241024T000000 role.

Select Assign Role -> Web identity -> GitHub organization.

AmazonS3FullAccess
AWSCodeDeployFullAccess

Add S3, AWSCodeDeploy permissions.

2.3 Create Amazon EC2

Fill in the name ec2.cheaper.001
Click Amazon Linux 2023 AMI
Click t3a.nano

Finally, click Launch instance to create EC2.

2.3.1 Associate Elastic IP address

Click on Elastic IPs
Click the Allocate Elastic IP Address button

Select the name ec2.paper.001 where EC2 has just been created
Select the default Private IP address
Click the Associate button

2.3.2 Amazon Route 53

Fill in the sub-domain name
Fill in the EC2's Private IP address
Click the save button

Successfully set up the static sub-domain name and IP address.

2.3.3 Add `AWS IAM` roles

Select Actions
Select Security
Select Modify IAM role

Add AWSCodeDeployService-EC2AccessCodeDeployRole-20241024T000000.

2.3.4 Install `CodeDeploy Agent` on `Amazon EC2`

Enter the Amazon EC2 terminal.

Select Connect button
Select EC2 Instance Connect tab
Select Connect button

Successfully log into the Amazon EC2 terminal.

sudo apt update
sudo yum install ruby
sudo apt install wget
cd /home/ec2-user
wget https://aws-codedeploy-us-east-2.s3.us-east-2.amazonaws.com/latest/instal
chmod +x ./install
sudo ./install auto

Install CodeDeploy Agent

Success, CodeDeploy Agent is running.

2.3.5 (Optional) Install `Git` on `Amazon EC2`

sudo yum install git-all
git clone https://{YOUR_GITHUB_SECRET_ID}@github.com/{YOUR_GITHUB_ORGANIZATION_NAME}/{YOUR_GITHUB_PROJECT_NAME}.git
git checkout .
git pull origin main
sudo chmod 777 -R PATH

Install git and pull the project to Amazon EC2.

2.3.6 (Optional) Install `NGINX`

sudo yum update
sudo yum install nginx -y
sudo service nginx start
sudo service nginx status

Install NGINX

sudo netstat -tunpl

Show Amazon EC2 listening ports. At this moment NGINX is on port :80.
The default home page of NGINX is in /var/www/html/index.html.

Ensure that Source and Destination are publicly accessible, set to 0.0.0.0/0.

2.3.7 Appspec.yml

Reference Articles:

Appspec.yml is used to indicate the codeDeploy procedure code.
Deployment is divided into 5 steps: (1) BeforeInstall -> (2) BeforeInstall -> (3) AfterInstall -> (4) ApplicationStart -> (5) ValidateService.

In the root directory, add ./appspec.yml.

version: 0.0
os: linux
files:
    - source: /
      destination: /usr/share/nginx/html
hooks:
    ApplicationStop:
    - location: scripts/application-stop.sh
      timeout: 300
      runas: root
    BeforeInstall:
    - location: scripts/before-install.sh
      timeout: 300
      runas: root
    AfterInstall:
    - location: scripts/after-install.sh
      timeout: 300
      runas: root
    ApplicationStart:
    - location: scripts/application-start.sh
      timeout: 300
      runas: root
    ValidateService:
    - location: scripts/validate-service.sh
      timeout: 300
      runas: root

Source is the root directory of the GitHub project.
Destination is the project pulled intoAmazon EC2.

In addition, a new ./scripts folder, in which there are 5 xxxxxxxx.sh respectively.

application-stop.sh
before-install.sh
after-install.sh
application-start.sh
validate-service.sh

There are 5 xxxxxxxx.sh in there, which are the 5 steps of codeDeploy.

(1) application-stop.sh

#!/bin/bash

Empty. There is no need to stop the application in this tutorial.

(2) before-install.sh

#!/bin/bash

Empty. There is no need to stop the application in this tutorial.

(3) after-install.sh

#!/bin/bash

sudo yum update
sudo yum install nginx -y

Install NGINX

(4) application-start.sh

#!/bin/bash

sudo service nginx start

restart NGINX

(5) validate-service.sh

#!/bin/bash

Empty. There is no need to stop the application in this tutorial.

2.3.8 Static Website Pages

Added ./icons folder, which shows the site image ic_alana_002_20241022_a.jpg.

Also, added index.html home page.

<html lang="en" data-bs-theme="dark">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=no">
<script src="https://cdnjs.cloudflare.com/ajax/libs/bootstrap/5.3.3/js/bootstrap.min.js" integrity="sha512-ykZ1QQr0Jy/4ZkvKuqWn4iF3lqPZyij9iRv6sGqLRdTPkY69YX6+7wvVGmsdBbiIfN/8OdsI7HABjvEok6ZopQ==" crossorigin="anonymous" referrerpolicy="no-referrer"></script>
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/bootstrap/5.3.3/css/bootstrap.min.css" integrity="sha512-jnSuA4Ss2PkkikSOLtYs8BlYIeeIK1h99ty4YfvRPAlzr377vr3CXDb7sb7eEEBYjDtcYj+AjBH3FLv5uSJuXg==" crossorigin="anonymous" referrerpolicy="no-referrer" />
<title>Alana Lam</title>
</head>
<body>
<div class="container">
  <div class="row">
  <div class="col-12 mt-4 text-center">
  <h1>CodeDeploy + Github Actions + EC2</h1>
  <img src="./icons/ic_alana_002_20241022_a.jpg" class="mt-4 rounded-circle" alt="Alana Lam" width="200" height="200">
  <h5 class="mt-4">Alana Lam (AWS Builder Community Manager, Hong Kong)</h5>
  </div>
  </div>
</div>
</body>
<html>

A simple static site with text and images.

If you have completed “2.3.5 Install GIT” and “2.3.6 Install NGINX”, you can type EC2 EIP or the domain name in your browser, to see the Static Website Pages.

2.4 Create `AWS CodeDeploy`

2.4.1 Create the `AWS CodeDeploy` application

Fill in the application name test.codeDeploy.001
Select EC2/On-premises
Select Create application button

2.4.2 Create `AWS CodeDeploy` Deployment Group

Select Create deployment group button

Fill in the Deployment group name test.deploymentGroup.001
Select the IAM role, AWSCodeDeployService-DepolyEC2Role-20241024T000000
Remove Enable load balancing, because this is the simplest DevOps pipeline case, so there is no need for additional AWS services

2.4.3 Create `AWS CodeDeploy` Deployment

Go to test.deploymentGroup.001
Select Create deployment button
First, Select My application is stored in GitHub

Fill GitHub token name
Fill in the Repository name, codedeploy.nginx.001
Fill in Commit ID
Select Create deployment button

2.4.4 Successful run of `AWS CodeDeploy`

Successfully run AWS codeDeploy

2.5 Create `GitHub Actions`

Reference Articles:

Why I switched from AWS CodePipeline to GitHub Actions

2.5.1 Create `GitHub Actions` workflow

Click New workflow button
Select set up a workflow yourself link
After writing the GitHub Actions command, click the Commit changes button

2.5.2 Configurate `GitHub Actions` secrets and variables

Select Settings Tab
Select Secrets and variables -> Actions Tab
Select Secrets Tab

2.5.3 Add `GitHub Actions secrets` variables

Add a new secrets variable with name IAMROLE_GITHUB_ARN
The value is the ARN of the IAM role arn:aws:iam::{xxxxxxxxx}:role/AWSCodeDeployService-GitAssumeRoleWithAction-20241024T000000
Click the Add secret button

2.5.4 Add `GitHub Actions variables`

Select Variables Tab
Add four of Actions Variables
Select New repository variable button

Variables Name	Value	Description
AWS_REGION	us-east-1	The default region is `US East (N. Virginia)`
CODEDEPLOY_APPLICATION_NAME	test.codeDeploy.001	2.4.1 Create the `AWS CodeDeploy` application
CODEDEPLOY_DEPLOYMENT_GROUP_NAME	test.deploymentGroup.001	2.4.2 Create `AWS CodeDeploy` Deployment Group
IAMROLE_GITHUB_SESSION_NAME	AWSGitAssumeRoleWithAction	2.2.3 AWSCodeDeployService-GitAssumeRoleWithAction-20241024T000000

2.5.5 Write `GitHub Actions` Code

.github/workflows/main.yml

name: Deploy

on:
  workflow_dispatch: {}

jobs:
  deploy:
    runs-on: ubuntu-latest
    environment: Prod
    permissions:
      id-token: write
      contents: read
    steps:
    - name: Git clone the repository
      uses: actions/checkout@v2

    - name: Configure AWS credentials
      uses: aws-actions/configure-aws-credentials@v4
      with:
        role-to-assume: ${{ secrets.IAMROLE_GITHUB_ARN }}
        role-session-name: ${{ vars.IAMROLE_GITHUB_SESSION_NAME }}
        aws-region: ${{ vars.AWS_REGION }}
    - run: |
        commit_hash=`git rev-parse HEAD`
        aws deploy create-deployment --application-name ${{ vars.CODEDEPLOY_APPLICATION_NAME }} --deployment-group-name ${{ vars.CODEDEPLOY_DEPLOYMENT_GROUP_NAME }} --github-location repository=${{ github.repository }},commitId=${{ github.sha }} --ignore-application-stop-failures

A basic version of the GitHub Actions Code.

2.5.6 Run GitHub Actions Code

Select Actions Tab
Select Deploy Tab
Select Run workflow button

2.5.7 Successfully running `GitHub Actions`

Successfully ran main.yml

4.0 Cost

Plan	USD
Monthly cost	$11.83
Total 12 months cost	$141.96

Overall, AWS's prices are quite competitive. The most important thing is that CodeDeploy is cheap, and the cost of using Amazon EC2 t4g.nano is very low, so AWS is a low-cost + efficient cloud service provider.

4.1 Detailed Estimate

Service	Monthly	First 12 months total (USD)
AWS CodeDeploy	$8.8	$105.6
Amazon EC2	$1.533	$18.4
Amazon Route 53	$0.4	$4.8
VPN Connection	$1.1	$13.2

5.0 Summary

GitHub Actions + CodeDepoly are powerful DevOps tools that fulfill the principle of “think big, take small steps” in a business environment.

To conclude, let's summarize the key points of this chapter:

5.1 Principles

The new “Macro Portfolio” system is to comply with the “Least Effort Principle”, which includes (1) agile development, and (2) agile deployment
The real issues were (1) the project took too long to deploy, and (2) automated deployment was not achieved
Success is due to the following: (1) Other departments want small features in small increments. (2) More simplicity means more understanding of the problem's root cause.

5.2 Action

Give the “Updated API Manual” to other departments to try before every Thursday
Simplicity is a good result of the Highest Standards because we performed (1) a “DIVE DEEP investigation” and (2) understanding the root cause of the problem

5.3 AWS DevOps

The development engineer commits the code via GitHub Push
GitHub Actions trigger workflows
IAMROLE_GITHUB_ARN authorizes access to AWS resources
GitHub Actions triggers AWS CodeDeploy
AWS CodeDeploy triggers deployment to Amazon EC2 instances
AWS CodeDeploy pulls Github resources and deploys to Amazon EC2 instances

5.4 `AWS IAM` （`CodeDeploy`, `EC2`, `Github`）

AWSCodeDeployService-EC2AccessCodeDeployRole-20241024T000000
AWSCodeDeployService-DepolyEC2Role-20241024T000000
AWSCodeDeployService-GitAssumeRoleWithAction-20241024T000000

5.5 `AWS CodeDeploy` （Appspec.yml）

BeforeInstall
BeforeInstall
AfterInstall
ApplicationStart
ValidateService

5.6 Cost

Monthly cost: $11.83 (USD)
Total 12 months cost: $141.96 (USD)

Postscript

On 14 December 2024, I attended the annual Amazon Greater China Community Gathering. I am very thankful to AWS for bringing me an unforgettable experience.

📷Shoot and 🎬Edit by Kenny Chan
Also, thanks to Smile (Lingxi) Lv - Developer Experience Advocacy Program Manager for supporting AWS Community Builder.

About Me

Kenny Chan, AWS Community Builder (Hong Kong), specialty of Fintech & Machine Learning

AWS DevOps+Q Agile Delivery of 16 Leadership Principles for the Financial Services Industry

Kenny Chan — Tue, 24 Dec 2024 07:37:29 +0000

1.0 Preface

I am currently working in a financial technology company that specializes in providing (1) financial trading data and (2) macro-asset allocation solutions. The company is developing a “Macro Portfolio” system to support other departments, such as Macroeconomic Analysis, Trading Systems, Risk Management, Financial Deep Learning, Cybersecurity Engineers, etc. The new “Macro Portfolio” system will be used by the company's financial services department to support the business of the financial services industry.

The new “Macro Portfolio” system is to comply with the “Least Effort Principle”, which includes (1) agile development and changes to respond to the volatility of the financial markets in the VOCA era, and (2) agile deployment of the project to reduce the time wasted in communication with other departments, and implementation of automation work.

However, the current development engineer (yes, that's me!) is uncomfortable developing a “macro-portfolio” system across multiple departments. Functional requirements from the macroeconomic analysis department, functional feedback from the risk management department, code changes from the financial deep learning department, and code review from the cybersecurity engineer department.

After two weeks of “DIVE DEEP investigation” and meetings with a “single leader” in each department, the real issues were (1) the project took too long to deploy, and (2) automated deployment was not achieved.

If we can release the functionality in GrayRelease first, and give the “Updated API Manual” to other departments to try before every Thursday, we can increase the overall project development speed by +30%. In other words, customers will be able to experience the new “Macro Portfolio” functionality in 10 days instead of 13 days.

Therefore, we decided to deploy the Macro Portfolio system using the AWS DevOps pipeline to accelerate the entire “Prototype->Development->Deployment->Use->Feedback->Modification” project lifecycle.

“Customer Obsession” is matter! -- Amazon 16 leadership principles

1.1 Goals of the AWS DevOps Pipeline

In October 2024, I took a week to read Akshay Kapoor's [AWS Senior Cloud Infrastructure Architect] AWS DevOps Simplified: Build a solid foundation in AWS to deliver enterprise-grade software solutions at scale. Then I understood what Raymond Tsang [AWS Senior Technical Trainer] told me at the Hong Kong Re: Invent Recap in February 2024 when he said: “There is no absolute right solution, so even if you use only a small portion of AWS services and the result is better, then that's a good solution.”

Now, I totally agree with Raymond Tsang [AWS Senior Technical Trainer]. This “AWS DevOps Pipeline” architecture was a success in results, and magically, only a few AWS services were used. For example, we didn't use Instance Auto Scaling, AWS ECS (Elastic Container Service), AWS ELB (Elastic Load Balancing), AWS CloudFormation, etc.

“Invent and Simplify” is a matter! -- Amazon 16 leadership principles

“Any damn fool can make it complex. it takes a genius to make it simple.” -- Ray Dalio, Principles

In my experience, success is due to the following: (1) Other departments want small features in small increments, not complete solutions. (2) More simplicity means more understanding of the problem's root cause. So (3) it's faster and more efficient to “Deliver Results”, even if it's just a small portion of AWS services.

Because the financial services industry is specialized, each department is responsible for different goals to help customers get value. Financial DevOps should not be a limitation or obstacle, but rather a way to better serve other departments and customers in different situations.

At the same time, I understood that “Simplify & Insist on the Highest Standards” is not a conflict. Although we simplified the whole project, simplicity is a good result of the Highest Standards because we performed (1) a “DIVE DEEP investigation” and (2) understanding the root cause of the problem.

Although you may not believe it, in emergency situations, we use Excel to calculate the Black-Scholes model, because Excel is the fastest and easiest way to solve emergency problems, and it is also the best for “Customer Obsession” and “Deliver Results”.

1.2 “AWS DevOps Pipeline” Architecture

The services used in the AWS DevOps pipeline:

GitHub Actions
AWS CodeDeploy
Amazon EC2
IAM

Walkthrough:

The development engineer commits the code via GitHub Push
GitHub Actions trigger workflows
IAMROLE_GITHUB_ARN authorizes access to AWS resources.
GitHub Actions triggers AWS CodeDeploy
AWS CodeDeploy triggers deployment to Amazon EC2 instances
AWS CodeDeploy pulls Github resources and deploys to Amazon EC2 instances

1.3 Optimization

This architecture is only suitable for Agile delivery and development environments that are currently “running small steps quickly”. If it is used in a production environment, we need to use Instance Auto Scaling, AWS ECS (Elastic Container Service), AWS ELB (Elastic Load Balancing), etc.

In addition, to make it easier for you to understand the “AWS DevOps pipeline” mechanism, in the following tutorials, Python and Backtrader are removed from the application layer, while just a simple Nginx and static web pages are used.

1.4 Applying GenAI Tools - Amazon Q in Financial Services DevOps

Amazon Q is a perfect GenAI Chatbot development tool.

In September 2024, I found an anomalous charge on my AWS bill, but I didn't know why EC2 EIP became a paid service.

So I asked Amazon Q and in just 10 seconds I knew the answer. The idle EIP is charged.

Select EC2 -> Elastic IP addresses -> Network & Security
Select Elastic IPs
Delete the idle EIPs

In addition, I used Amazon Q to learn about AWS DevOps. The following is my experience with Amazon Q while applying the knowledge from AWS Certified Machine Learning - Specialty Certification exam.

Since AWS Pipeline is an AWS-Centric service, integrating GitHub Actions does not meet the Least Effort Principle.

Therefore, AWS Pipeline is not the best approach.

I know that AWS Pipeline is AWS-Centric Orchestrates and that it consists of CodeBuild and CodeDeploy.

I asked Amazon Q, knowing that CodeDeploy was the AWS service I needed and would be the best Least Effort Principle solution with GitHub Actions.

Reference Articles:

Simplify Amazon EKS Deployments with GitHub Actions and AWS CodeBuild

After I read the tutorial on the AWS DevOps blog, it is already very similar to my solution although it is AWS EKS.

Through Amazon Q, I quickly understand the differences and similarities of each AWS service and apply it more productively in my daily work. Therefore, I highly recommend using AI tools for productivity.

1.5 Summary

I shared the current situation in the financial services industry and then applied the AWS 16 Leadership Principles, Akshay Kapoor and Raymond Tsang's insights on solving business and technical pain points through AWS cloud services.

Finally, summarize the key points of this chapter:

1.5.1 Principles

The new “Macro Portfolio” system is to comply with the “Least Effort Principle”, which includes (1) agile development, and (2) agile deployment
The real issues were (1) the project took too long to deploy, and (2) automated deployment was not achieved
Success is due to the following: (1) Other departments want small features in small increments. (2) More simplicity means more understanding of the problem's root cause.

1.5.2 Action

Give the “Updated API Manual” to other departments to try before every Thursday
Simplicity is a good result of the Highest Standards because we performed (1) a “DIVE DEEP investigation” and (2) understanding the root cause of the problem

1.5.3 AWS DevOps

The development engineer commits the code via GitHub Push
GitHub Actions trigger workflows
IAMROLE_GITHUB_ARN authorizes access to AWS resources
GitHub Actions triggers AWS CodeDeploy
AWS CodeDeploy triggers deployment to Amazon EC2 instances
AWS CodeDeploy pulls Github resources and deploys to Amazon EC2 instances

In the next chapter, I'll share (1) building an AWS DevOps pipeline, and (2) the estimated cost of AWS cloud services. I hope you all grow together in the AWS community.

About Me

Kenny Chan, AWS Community Builder (Hong Kong), specialty of Fintech & Machine Learning

中国社区日Deep Talk：与八位IT从业人的深度对话要点总结 [Updated: 2024-11-12]

Kenny Chan — Tue, 16 Jul 2024 05:14:50 +0000

Danny Chan speaking AI topic in AWS User Group (Shen Zhen, China) on 2024-07-06

More Photos of AWS COMMUNITY DAY in Shen Zhen

直播回放 of AWS COMMUNITY DAY in Shen Zhen

1. 传统IT拥抱GenAI浪潮 - 惠州穆老板

「创办人的领导力分享」一家IT公司的商业战略

「轻资产+周期快+现金流稳定」的IT公司，利用GenAI开拓更大的应用场景
「轻资产+周期快+现金流稳定」有可复制性和易推广的两大商业优势。因可在完成一个项目后，再贩售给其他同类型行业的买家，靠行业间「口碑营销」易吸引新买家，从而积累成「行业解决方案」。
这种行业解决方案有「业务短+领域知识浅」的两大特性，如债务公司，牙医公司等行业。
AI场景可开拓在长业务的其中一个痛点环节上。例如，因中国很多中国中小企工厂未做到全面数字化管理，所以存在「制造」和「销售」的断裂。因此，AI可只切入长工厂生产环节的其中一个痛点，比如是只协助销售人员预测下季的出货量等。
在团队推动共享知识。因只有提升团队的整体知识水平，才可以拥抱AI浪潮开拓新业务，所以很鼓励员工多参加行业分享活动，如AWS User Group等。
员工掌握时间的能力很重要。因每个任务的重要性和耗时性都不一样，所以要先选择最重要的任务，平均地分配每个环节时间，从而交出完整度高的结果。

2. 家庭幼教赋能AI - Miss Xi

「家庭幼教的AI场景分享」zero code的AI产品仍是领域专家的工具

目前AI产品工具仍然不好使用。即使百度和腾讯推出面向大众的zero code AI产品，但对于想造出拥有专业领域知识的AI产品仍很难。
用户界面不好使用。因家庭幼教是多文书类的场景，虽然很合适用AI助力，但在UI输入资料的过程仍然不友好。
模型效果难看明白。因爲zero code AI产品没有可视化的模型表现页面，加上也看不懂模型数据，所以大众搞不明白怎么用。

3. 金融数据分析BI到AI产品 - Miss Hai

「金融数据人的火花分享」当金融BI遇上香港传统金融

Miss Hai（问）：传统金融是因追赶AI浪潮，而被动接受AI转型？
Mr Kuang（答）：作为传统金融人，我们在十多年前就拥抱AI。但是能应用AI的金融场景是稀罕的，所以只有到近两年AI技术完善了，才能打通金融场景落地。

Miss Hai（问）：那么AI可以应用在哪些金融场景。
Mr Kuang（答）：主要是在文字类的金融场景，如新闻情绪分析，研报态度分析。但这些金融AI场景已经存在很多年，只是近两年AI技术变得更完善，如有65万个参数的大模型，才能实现十年前的AI幻想场景。

4. 香港大学博士研究团队成员 - Mr Xu

「实验室项目的落地分享」跨界的商业协调人

要知道领域的「商业逻辑」，但研究和开发的流程该留给领域专家。因爲自己怎么学也干不过领域专家，反而浪费自己的学习时间成本。
更多是协调资源，但首先要理解领域的「商业逻辑」，才可以更好地协调资源。
曾做过AI，金融，医疗等项目。采用集团思维，虽然三个是完全独立的项目，但享受了「品牌口碑」优势。

5. 副业是连续创业者 - Mr Li

「work life balance分享」正职打工人和副业创业者，双面人生的平衡

中国的创投比赛更着重商业逻辑，即使是公益类的赛道，仍然看重商业逻辑和应用场景。
香港的创投比赛更容许创新性，但中国因注重商业逻辑而影响创新性。
正职是在一家外资的全球物流公司工作，可以远程工作，所以有更多时间成爲创业者。
有做过AI小说生成视频项目，AI聊天机械人项目。对于金融聊天机械人，只集中在应用场景开发，而专业的金融领域知识交给专业金融人来负责。

6. AI创业家 - Mr Jack

「AI产品商业化分享」创投基金和投资人是两种不同玩法

创投基金只是帮助第一阶段的产品落地，但对商业变现的扶助是不足够的，所以更多是学习阶段。
优先与投资人建立好关系。因投资人是最大VIP用户，所以要优先关注好投资人的看点，与投资人建立好关系。

7. 高级解决方案工程师 - Mr 张

「GenAI概念项目技术分享」多个AI agent的多轮对话项目

需要一个主持人AI agent。因爲多个AI agent有以下问题：1. 没有共同目标。 2. 胡乱次序发言。 3. 互相不知大家讲了什么。
因此主持人agent隐藏在幕后的作用：1. 定义每个agent身份。 2. 预输入knowledge base给大家。3. 定义共同目标。 4. 控制和协调发言流程 5. 广播大家的对话。
但仍有一些限制：1. 如果AI agent提出错误内容，主持人也无法修正，导致整个流程失败。

8. 市场策划人 - Miss Nan

「行业经验分享」游戏发布，抖音电商

「游戏发布」是指帮助对接工作室和市场营销资源。
无论小中大的游戏工作室，都需要「游戏发布」中间商来帮助对接「市场营销」。「游戏发布」价值点就是人脉资源，而盈利点是收X%提成。
「抖音电商」本质是卖流量转化率。如XX游戏花7000万投放抖音广告，就是买流量转化率，目标是引导用户经抖音广告去行动，比如是下载游戏，购买商品等。
「抖音电商」有大量的用户数据和营运数据，所以需分析哪些流量最有效率和最值钱，和商业预测怎样才能提升盈利点。

9. 前端开发者转型市场策划 - Miss Ya

「女性职场环境分享」女性开发者的困局

女性开发者仍存在被不公平对待的现像。
第一，来自前端的内卷文化。因爲前端的开发者人数多+任务复杂，所以重要的任务都交给男性开发者，导致女性开发者没有晋升机会。
第二，女性的职场生命短。因爲在中国观念中，已婚女性主要是家庭主妇，就算重返职场也被视爲缺少工作经验。
第三，女性在职场的不稳定性。因女性有产假，生育假等，令主管不想聘请有可能令职位空缺的女性开发者。
从前端开发者转型爲市场推广的原因，除了有行业内卷文化，还有中国职场对女性不友好。

开发者生态圈观点：

1. 怎么看996文化？
惠州穆老板：从来没有要求员工加班，更多的是希望员工有优秀的时间控制能力，更有效率地完成工作。比如有些是公司内部的工作，可以重新排序重要性，而非通过加班来解决。

Mr 张：如果有项目交付，偶尔是有限度的加班。最重要还是项目的完成度。比如展览，就需要提早做好现场布置，硬件调试，与团队对接等。

2. 怎么看八股文文化？
惠州穆老板：不会考核八股文，更多是考核实际的工作能力，比如是会多少开发语言和工具，有多少开发经验等。

Mr Li：打工人有需要背八股文，因同行竞争大，所以只能靠八股文争拼。

Editor

Danny Chan, AWS community builder (Hong Kong), specialty of FSI and Serverless

Kenny Chan, AWS community builder (Hong Kong), specialty of FSI and Machine Learning

Most three important Q&A of Trading Strategy Deployment for AWS SageMaker

Kenny Chan — Fri, 26 Jan 2024 16:48:17 +0000

Introduction

I am extremely delighted to have participated in the AWS re:Invent re:Cap event held in Hong Kong, which provided me with exposure to the latest AI solutions offered by AWS.

In my previous article, although I discussed deploying deep learning models in production using EC2, such a solution is only suitable for my personal use case, which can be found in the article "Machine Learning Trading Strategy Best Practices for AWS SageMaker".

In this article, I will first discuss the advantages of deploying models in production using SageMaker after training them locally. I would like to express my gratitude to Raymond Tsang for providing valuable insights.

Next, I will delve into the benefits of training models using SageMaker as opposed to local training. I would like to thank Yanwei CUI for sharing their insights.

Lastly, I will explain a more efficient trading strategy architecture, with special thanks to Wing So for their valuable input.

1. The Benefits of Deploying Models in Production with SageMaker

The greatest advantage of SageMaker lies in its data security, auto scaling, and container deployment capabilities. If high data security, handling sudden traffic spikes, and agile development processes are required, leveraging these advantages of SageMaker can significantly accelerate development and deployment timelines.

However, after training models locally, can one deploy them in production using SageMaker? In other words, is it possible to utilize only specific functionalities of SageMaker?

Answer: Yes, it is possible to use only certain functionalities of SageMaker.

In the case of my use case, "Alice's Intraday Futures Trading Strategy," which is a daily trading strategy model with fixed trading times and a predictable number of requests, the model is susceptible to market sentiment and unexpected news events, necessitating monthly model updates.

In such a scenario, deploying the model in a production environment using SageMaker offers the following advantages:

SageMaker allows for container deployment, making it easier to manage custom inference code within the deployment image.
SageMaker's endpoint supports version iterations, facilitating agile development processes.
SageMaker supports multi-model deployment in a single endpoint, enabling easier management of multiple model interfaces.

While local model training is preferred in my use case, there are still advantages to using SageMaker for model training.

2. The Advantages of Training Models with SageMaker

If there are two RTX3080 graphics cards available on the local server, is there still a need to use AWS SageMaker for training models? In other words, can one replace the pay-as-you-go model training of SageMaker with a one-time higher fixed cost?

Answer: Yes, it is possible. However, if one wishes to avoid the time-consuming process of hardware deployment or simply desires to utilize higher-end hardware for a shorter duration, training models using SageMaker is more suitable.

Furthermore, SageMaker optimizes data-batch processing and floating-point operations to accelerate model training speed.

In the case of my use case, "Diana's Medium-Term Quarterly Trading Strategy," which involves multi-asset trading in four major markets (US stocks, Hong Kong stocks, US bonds, and USD currency), the optimized data-batch processing of SageMaker can be utilized for the four main markets.

Additionally, the optimized floating-point operations of SageMaker can be applied to the three core technical indicators within the model (high dividend stocks, low volatility, and capital accumulation).

Therefore, gaming graphics cards have limitations when it comes to model training.

3. A More Efficient Trading Strategy Architecture

Whether using EC2 or SageMaker container deployment, both options serve to expedite development time. However, considering the overall efficiency of the trading system, two factors need to be considered: streaming data processing and the layer at which computations are performed.

The key to achieving higher efficiency lies in the Queue layer.

After the Data Provider delivers streaming data, the Queue distributes the data to the Application while simultaneously storing the streaming data in a database. This reduces latency and improves overall efficiency.

Furthermore, performing computations at the Queue layer for the technical indicators used by all Applications prevents redundant calculations and enhances overall efficiency.

However, further investigation is required to determine which Queue framework to use.

Summary

The theme of AWS re:Invent re:Cap, "Gen AI," was a captivating event. There were many intriguing segments, such as the "Deep Dive Lounge," "Lighting Talk," and "Game Jam," which provided delightful surprises.

Deep Dive Lounge, Wing So.

More importantly, numerous AWS solution architects have contributed to the advancement of my trading endeavors, offering lower-cost solutions and improved computational efficiency. Lastly, I would like to express my special thanks to Raymond Tsang, Yanwei CUI, and Wing So for their invaluable assistance.

Machine Learning Trading Strategy Best Practices for AWS SageMaker

Kenny Chan — Sun, 14 Jan 2024 12:31:04 +0000

Introduction

In my previous articles, I used two different trading strategies to explain the best practices of batch-transform and real-time endpoints, as well as the reasons for using EC2. These articles can be referred to as "Even though Sagemaker provides various benefits, why do I still use EC2?" and "Why Choose Sagemaker Despite Having a Local Server with RTX3080?".

In this article, I will first demonstrate the complete architecture of SageMaker.

Then, I will explain the reasons for using Multi-Modal-Single-Container + Microservices and not using Application Load Balancer.

Finally, I will use two different trading strategies to explain the best practices of data parallelism and model parallelism in advanced training models.

Architecture Overview

Local Development Environment

CUDA 11.5 and Nvidia-container-toolkit for local model training.
jupyter/tensorflow-notebook for local development environment, with libraries required for Sagemaker[local], Backtrader, and Monitor Web UI installed in the image.

Supported AWS services

Sagemaker prebuilt images for pulling images to the local development environment for local model training and testing.
S3 Bucket for storing datasets and models.
CodePiplinefor deploying projects on Github to EC2 production environment.

EC2

Custom Production Container with libraries required for Sagemaker, Backtrader, and Monitor Web UI.
Monitor Web UI for presenting the trading performance of the model in graphical form, providing :80 to Trader and Asset Portfolio Manager.
Server Image for deploying models using Sagemaker prebuilt image, providing :8080 to business user.

Managed AWS Services

RDS for storing model results. Monitor Web UI in EC2 retrieves the data from RDS and presents the trading performance in graphical form.
CloudWatch for monitoring the computation and storage of EC2, RDS, and S3 Bucket.
IAM for helping jupyter/tensorflow-notebook in local development environment to access Sagemaker prebuilt images and S3 Bucket.

Why not use `Application Load Balancer` and instead create `Multi-Modal-Single-Container` + `Microservices` on EC2 to handle errors?

Application Load Balancer is a remarkable service. In fact, it can also be used to handle errors. However, in the case of trading strategies, I would choose to handle errors with Multi-Modal-Single-Container + Microservices.

Here are my three error handling methods:

The goal of the following three error handling methods is to flexibly reduce hardware resource requirements.

1.Switch to Smallest Model

There are two trading strategies (Diana's medium-term quarterly trading strategy and Alice's intraday futures trading strategy). Each trading strategy has two versions of the model, where the Biggest Model provides high accuracy but requires high hardware resources. On the contrary, the Smallest Model provides low accuracy but requires low hardware resources.

If the server is in a high computational state, switching to the Smallest Model can reduce the hardware resource requirements and keep the application running smoothly.

2. Response caching results

When the same business user uses the application frequently, returning cached data can avoid overloading hardware resources.

3. Delayed Response time

When hardware resources are overloaded, delaying the response time can release the hardware resources.

Advantages of `Multi-Modal-Single-Container` + `Microservices`

Here are my examples of trading strategies to explain the reasons for using Multi-Modal-Single-Container + Microservices.

1.Trading strategies have high fault tolerance

Both trading strategies anticipate reduced profits due to slippage during trading. This design with high fault tolerance can accommodate various hardware issues, such as switching to the Smallest Model, response caching results, and delayed response time.

Additionally, it can handle errors from market makers, such as delayed quotes, partial executions, and wide bid-ask spreads.

2. Shared hardware resources

The frequency and time of use of two trading strategies are different, allowing for full utilization of idle hardware resources.

3. Deployment of trading strategies in different regions

Diana's medium-term quarterly trading strategy targets global assets. By deploying trading strategies independently in Hong Kong and the United States, the latency can be reduced.

Furthermore, if the hardware in Hong Kong completely stops working, the hardware in the United States can be used to hedge the risk by purchasing short options of overseas ETF.

Best Practices of Data Parallelism and Model Parallelism in Advanced Training Models

Sagemaker provides remarkable advanced training methods: Data parallelism and Model parallelism. I will use two different trading strategies to explain the best practices of data parallelism and model parallelism in advanced training models.

Data parallelism

Model parallelism

Model Parallelism: A simple method of model parallelism is to explicitly assign layers of the model onto different devices.

Data Parallelism: Each individual training process has a copy of the global model but trains it on a unique slice of data in parallel with others.

-- Accelerate Deep Learning Workloads with Amazon SageMaker, chapter10

In simple terms, if the data can be divided into small groups, Data parallelism is used. If the model can be divided into small groups, Model parallelism is used.

Alice's intraday futures trading strategy

The intraday trading strategy mainly uses a few key indicators to train the model, providing entry and exit points. Therefore, the data samples are large.

When the data sample is large and the model has only a few algorithms, Data parallelism should be used to train the model. This allows the data set to be split and computed on different GPUs.

distribution = { 
    "smdistributed": { 
        "dataparallel": {
            "enabled": True, 
            "custom_mpi_options": "-verbose -x NCCL_DEBUG=VERSION"
        }
    }
}

3_SDP_finetuning_pytorch_models.ipynb

Sagemaker provides remarkable advanced training methods. By setting the distribution parameter, Data parallelism can be used to train the model.

Diana's Medium-Term Quarterly Trading Strategy

The macro trading strategy mainly uses dozens of key indicators to provide overseas asset allocation forecasts. The minimum data set is 8 years (2 bull and bear cycles) of hourly snapshot data.

When the main algorithms can be split into small groups, Model parallelism is used to train the model. This allows the model tensor to be computed in batches on different GPUs.

distribution={
    "smdistributed": {
        "modelparallel": {
            "enabled":True,
            "parameters": {
                "microbatches": 8,
                "placement_strategy": "cluster",
                "pipeline": "interleaved",
                "optimize": "speed", 
                "partitions": 2,
                "auto_partition": True,
                "ddp": True,
            }
        }
    },
    "mpi": {
          "enabled": True,
          "processes_per_host": 1,
          "custom_mpi_options": "-verbose -x orte_base_help_aggregate=0" 
    },
},

3_SDP_finetuning_pytorch_models.ipynb

Similarly, by setting the distribution parameter, Model parallelism can be used to train the model.

Conclusion

AWS provides convenient solutions for the financial industry. Sagemaker seamlessly integrates deep learning workflow into production environments. Additionally, Sagemaker offers surprising features to accelerate development. I will continue to learn about new AWS products and share examples of AWS services in finance and trading.

Forem: Kenny Chan

When AI Starts to Understand Market Narratives: How the AWS Financial Services Symposium Is Rethinking Event‑Driven Trading

当AI开始读懂市场叙事：AWS Financial Services Symposium重构事件驱动交易的底层逻辑

Key Takeaways

① Turning unstructured information into signals.

② Fixing long-standing problems in discretionary macro trading.

③ Validating execution at the trading desk level.

Stories of how macro traders earn in the market?

AWS FSI Partner Sponsors (Golden Tier)

① Sensors Data (神策数据)

② IntSig (合合信息)

③ InsureMO (易保云)

④ DBAPPSecurity (安恒信息)

⑤ Deloitte Consulting (德勤咨询)

Portfolio Management for Macro Hedging (Updated:2026-01-09)

宏觀對沖的組合管理 Portfolio Management for Macro Hedging

核心觀點

Core Idea

1.0 Quarterly Correction Assets

2.0 90-Day Rolling Window Correction Assets

3.0 蒙地卡羅回測宏觀對沖的組合管理 Monte Carlo Backtest for Macro Hedging Portfolio Management

4.0 宏觀對沖實務 Macro Hedging in Practice

4.1 第一不平等：沒效率的對沖 First Inequality: Inefficient Hedging

4.2 第二不平等：時間等於風險 Second Inequality: Time Equals Risk

4.3 “擇時+配置”兩大核心，在美國市場既有效率對沖，也有時間收益，而減少風險。 Timing + Allocation — make U.S. markets efficient for hedging, while also giving time-based returns and reducing risk.

4.4 每個“資產、時間、收益、風險”皆不平等。 Every asset, time, return, and risk is unequal.

4.5 三大支柱：宏觀面、技術面、交易系統。蒙地卡羅几何收益路徑優化“宏觀面”。Three pillars: macro view, technical view, and trading system. Use Monte Carlo simulation to optimize geometric return paths for the macro side.

4.6 杠杆+複利，是宏觀對沖的兩大隱藏優勢。Leverage and compounding are two hidden strengths of macro hedging.

4.7 宏觀對沖交易實務。Macro hedging in practice.

5.0 交易執行實務。①目標、②資產、③紀律、④進步、⑤交易。Trading execution essentials: ① Goals, ② Assets, ③ Discipline, ④ Growth, ⑤ Trades.

5.1 資產池 Asset Pool

5.2 宏觀面 Macro Analysis

5.3 技術面 Technical Analysis

5.4 交易系統：倉位 Trading System: Position Sizing

5.5 交易系統：買入、止盈、止損 Trading System: Entry, Take-Profit, Stop-Loss

6.0 宏觀交易機構跟別人不同。我們聚焦在“交易風險形態”。Macro trading firms are different. We focus on the shape of trading risk.

6.1 宏觀交易員SOP3.5

6.2 宏觀交易員SOP3.5、輸入、輸出、可控指标KPI

6.3 宏觀交易員SOP3.5、AI自動化工具、交易員工具、總占比%、自動%、手動%

7.0 交易以外的工作

7.1.1 個人品牌Pitching。量化AI、宏觀交易員、前台交易桌。QuantAI、Macro Trader、Broker 。

7.1.2 個人品牌分成三個階後：定位，成長，精品

7.1.3 我的品牌矩陣。IC精品行業家，提供戰略價值，跨領域協調，且有專業技術深度。走輕管理風格。

7.1.4 從人的角度去想，他是如何想，如何做。

7.2 AI生產力，提高跨領域宏觀交易收益 AI Productivity to Enhance Cross-Domain Macro Trading Returns

About Me

中港台「新经济，新金融」路演有感 Reflections on the “New Economy, New Finance” Roadshow

核心观点

延伸阅读

聰明交易，迅速決策：以 AWS AI 洞察市場、發掘 Alpha 機會

台湾有感

精彩语录

金融FSI行业篇

量化对沖基金篇

我学了什么？

亚马逊领导力+桥水原则，基於现实的正确思考工具

①每人眼中都有一个现实，因此每个“现实”各有不同。

②现实是指“务实”，即向现实发动时间钳形攻击。

③错误地理解现实，走向失败的极端思考。

④朱熔基，“实务派”银行家，最佳实践“桥水原则”正确地思考。

Postscript

About Me

Macro Investments: Accelerating Market Data Decisioning with Bedrock & Q Developer [Tech for Trading series]

AWS Community Builder at AWS Summit Hong Kong 2025

My reflection at the AWS Summit Hong Kong 2025:

Marco Investments: Accelerating Market Data Decisioning with Bedrock on AWS [lv200,FSI,AI]

Postscript

Why did I choose AWS to grow my trading journey?

AWS x Macro Trader

Pure alpha x Absolute risk control

Excellent Writing Communication: Commander Intent, Amazon Narrative, and Trading Strategy Notes

What is Commander Intent?

How can ⭐Amazon culture⭐ be leveraged to reinforce trading desk performance?

About Me

为什么我选择AWS来提升我的交易生涯？（中文版摘要）

AWS x Macro Trader

更新2025-07-28：

纯阿尔法收益 x 绝对风险可控

优秀写作沟通： 指挥官意图 和 亚马逊叙事 和 交易策略笔记

如何利用⭐亚马逊文化⭐，强化交易桌绩效？

优秀写作沟通：指挥官意图和亚马逊叙事和交易策略笔记

2.3.3 Add `AWS IAM` roles

2.3.4 Install `CodeDeploy Agent` on `Amazon EC2`

2.3.5 (Optional) Install `Git` on `Amazon EC2`

2.3.6 (Optional) Install `NGINX`

2.4 Create `AWS CodeDeploy`

2.4.1 Create the `AWS CodeDeploy` application

2.4.2 Create `AWS CodeDeploy` Deployment Group

2.4.3 Create `AWS CodeDeploy` Deployment

2.4.4 Successful run of `AWS CodeDeploy`

2.5 Create `GitHub Actions`

2.5.1 Create `GitHub Actions` workflow

2.5.2 Configurate `GitHub Actions` secrets and variables

2.5.3 Add `GitHub Actions secrets` variables

2.5.4 Add `GitHub Actions variables`

2.5.5 Write `GitHub Actions` Code

2.5.7 Successfully running `GitHub Actions`