How I Read a Chart: The JJ Analysis Process, Step by Step 내가 차트를 읽는 방법 — JJ Analysis 분석 프로세스 공개

Why Transparency Matters

The most dangerous analysis is the kind where you can’t see the inputs. A chart with a conclusion but no methodology is just an assertion with graphics. This post lays out exactly how I build every chart and analysis you see on this site — so you can verify it, critique it, or build on it.

Step 1: Define the Question First

Before touching data, I write out the specific question I’m trying to answer.

Not “let me look at Bitcoin” — that’s exploration without purpose. The question should be:

• “Is Bitcoin currently above or below its historical post-halving pattern at month 14?”
• “Has the S&P 500 ever traded at this valuation level without being followed by a recession within 18 months?”
• “When the yield curve un-inverted in prior cycles, how long until the recession began?”

A specific question determines what data I need, what time range is relevant, and what would constitute a meaningful answer. Without a question, data exploration produces confirmation of whatever you already believe.

Step 2: Source the Data

Every chart I publish uses primary source data. For equities:

• Yahoo Finance / Bloomberg for price data
• FRED (Federal Reserve Economic Data) for macroeconomic series (CPI, unemployment, yield curve, M2)
• CoinGecko / Glassnode for crypto price and on-chain data

I document the source and date of retrieval for every dataset. This is not bureaucracy — it’s reproducibility. If my chart shows something different from another analyst’s chart on the same data, the source comparison is the starting point for resolving the discrepancy.

Step 3: Log-Transform the Price Series

For any price series spanning more than 2–3 years, I immediately log-transform. This is not optional — it’s structural.

The log transformation converts percentage changes into equal visual distances. A 50% drawdown in 1974 and a 50% drawdown in 2008 look identical on a log chart. On a linear chart, the 2008 drawdown is visually 10x larger because the nominal price was higher.

All regression fitting (Step 5) is done on log-transformed data.

Step 4: Establish the Time Range

Time range choice dramatically affects what a chart shows. A common manipulation technique in financial media is selecting a start date that makes one interpretation look better than another.

My default: use the maximum available data. For the S&P 500, that means 1927. For Bitcoin, 2013 (when price data becomes meaningful). For specific macro series, whatever FRED provides.

When I use a shorter time range (e.g., post-2013 for Bitcoin), I state why: either data quality before that point is suspect, or the asset fundamentally changed (pre-Satoshi Bitcoin ≠ institutional-era Bitcoin).

Step 5: Fit the Trend Line

For long-term analysis, I fit a log-linear regression to the price series. This is a simple ordinary least squares (OLS) regression of log(price) on time.

The resulting trend line represents the constant growth rate that best explains the long-run price trajectory. Deviations from this line — measured in standard deviations — are the primary analytical output.

For channel construction, I identify:

• Upper boundary: drawn through the highest notable peaks (not all-time highest, but significant cycle tops)
• Lower boundary: drawn through the lowest notable troughs

The distinction between a data-fit trend line and a hand-drawn one: the trend line is reproducible by anyone with the same dataset. A hand-drawn line depends on artistic judgment.

Step 6: Measure the Current Position

Given the trend line and channel boundaries, where does current price sit?

I express this as:

• Percentage above/below trend (e.g., “currently 45% above the log-linear trend”)
• Channel percentile (e.g., “currently in the 75th percentile of the channel range”)
• Standard deviations from trend (e.g., “+1.2 standard deviations above trend”)

All three expressions communicate the same thing differently. The percentage form is most intuitive for non-technical readers. The standard deviation form is most useful for comparing across different assets (e.g., “S&P 500 at +1.2 SD” vs. “Bitcoin at +0.8 SD” allows cross-asset comparison).

Step 7: Apply Cycle Context

A price position means more when combined with cycle context. I overlay:

• For equity analysis: NBER recession bands, Fed rate cycle phases, major events
• For Bitcoin analysis: halving dates, post-halving day count, prior cycle overlays

Cycle overlays don’t predict the future — they contextualize the present. “Currently at +1.2 SD above trend, month 14 of the halving cycle” is a more complete statement than either alone.

Step 8: State the Interpretation and the Falsification Condition

Every analysis ends with:

1. What the data suggests
2. What would need to be true for that interpretation to be wrong

The second part is as important as the first. An interpretation without a falsification condition is a belief, not an analysis.

Example:

• “The data suggests we are in a mid-cycle correction, not a cycle top.”
• “This interpretation is wrong if: MVRV exceeds 6, or the 200WMA breaks, or the halving cycle reaches month 24+ without a new ATH.”

These conditions are checkable. If they occur, the interpretation is updated. The data leads; the narrative follows.

The Tools I Use

• Python (pandas, numpy, matplotlib) for data processing and chart generation
• FRED API for macroeconomic data retrieval
• CoinGecko API for crypto price data
• Google Sheets for quick data exploration and client-ready tables

Everything is reproducible from public data sources. Nothing I publish requires proprietary data or paid terminals.

An Invitation

If you want to replicate any analysis, verify any chart, or challenge any conclusion on this site — I welcome it. Transparent analysis isn’t just a methodology choice; it’s a philosophical commitment.

Send me your chart. Show me where I’m wrong. That’s how the analysis gets better.

All content represents personal research and opinion. Not investment advice.