lyndonkl/claude/payout-policy-analysis

Payout policy analysis

Three questions, in order. How much cash did the firm return? How much could it have returned? And do you trust management with the difference?

The first two are arithmetic and the script does them. The third is the judgment the whole analysis exists to support, and the script only scores the evidence for it.

The single most common error in this work is looking at dividends alone. Buybacks are roughly 60% of cash returned in the US. A firm with a 19% dividend payout ratio can be returning 83% of its earnings, and a firm with an 18% payout ratio can be unable to afford it. Add buybacks on both sides of every comparison before drawing a conclusion.

The script

resources/payout.py — pure standard library, no installation needed. Every subcommand takes JSON on stdin (or --in FILE) and prints JSON.

python3 resources/payout.py <subcommand> --example    # show input shape
python3 resources/payout.py <subcommand> --in payload.json
python3 resources/payout.py selftest                  # verify the engine

The bundled examples are Damodaran's own cases, so --example | <subcommand> reproduces his worked numbers. Run selftest after editing anything.

Work it in this order

Payout analysis progress:
- [ ] 1. Measure what was actually returned, dividends plus buybacks
- [ ] 2. Compute FCFE over 5 years, all three variants
- [ ] 3. Score the trust evidence
- [ ] 4. Read the matrix quadrant
- [ ] 5. Project forward and test sustainability
- [ ] 6. Cross-check against peers and the market

1. What was returned

Pull dividends paid on common stock and stock repurchases from the financing section of the cash flow statement, for each of the last five years. Report the split as well as the total, because the split tells you whether management deliberately kept flexibility.

Net buybacks against equity issuance where stock compensation is large. Gross repurchases at such a firm partly offset dilution rather than returning cash to anyone.

2. FCFE, in three variants

python3 resources/payout.py fcfe-history --in disney.json

Each year needs net_income, depreciation, capex (including acquisitions), change_in_noncash_wc, net_debt_issued, dividends and buybacks, most recent first. Add preferred_dividends where they exist.

The three variants differ only in how reinvestment is assumed to be financed:

Reinvestment       = (Cap Ex − Depreciation) + ΔNon-cash Working Capital
FCFE (pre-debt)    = Net Income − Reinvestment
FCFE (actual debt) = FCFE (pre-debt) + Net Debt Issued
FCFE (target DR)   = Net Income − Reinvestment × (1 − DR)

Lead with the target-debt-ratio variant. Actual-debt FCFE counts one-off borrowing as payout capacity, which flatters an acquisitive, debt-funded firm. Tata Motors returned 9% of actual-debt FCFE and 157% of target-ratio FCFE over the same five years. Only the second number is sustainable capacity.

Use the actual-debt variant when debt policy is stable and deliberate. Say which one your conclusion rests on. The engine sets variants_agree to false when the three disagree on whether the firm is an accumulator or an overpayer, and that disagreement is itself a finding worth reporting.

Setting the debt ratio. Use the current market debt-to-capital ratio unless you are deliberately moving the firm to a target. If you use a target, use the same one in the forward projection, and say you changed it. Get the market value of debt from cost-of-capital-toolkit's mv-debt.

Five years, not one. A single year is dominated by lumpy cap ex and borrowing. The engine accepts one to ten years and refuses anything outside that.

Read the output at comparison. Each variant carries the aggregate FCFE, the surplus or deficit in currency, cash returned as a percentage of FCFE, and an accumulator or overpayer label. Where FCFE is negative the percentage is reported as null with a reason, because a negative denominator makes it meaningless. Read the surplus in currency instead.

3. The trust evidence

python3 resources/payout.py trust --in disney-returns.json

Trust is not a feeling. It is measured two ways, and they disagree often enough that reporting one alone is a mistake.

ROE_t             = Net Income_t / Book Equity_t
Required Return_t = Rf_t + β × (Rm_t − Rf_t)
Excess ROE_t      = ROE_t − Required Return_t
Jensen's alpha_t  = Stock Return_t − Required Return_t

The project measure asks whether the investments earned their cost of equity. The market measure asks whether the stock beat the same bar. Disney over the five years to 2013 shows average ROE of 3.02% against an average required return of 11.22%, yet the stock beat its required return by 6.42% a year. On the accounting measure management destroyed value; on the market measure it created value.

When they conflict, weight the project measure for the payout decision. A rising market lifts alpha at firms whose managers did nothing well. Say which measure you relied on.

Two cautions on the inputs. Book equity is distorted by buybacks, write-offs and acquisition accounting, so a low ROE at an acquisitive firm may be an artifact. And use each year's own risk-free rate and market return, not a single long-run average.

average_roe_method defaults to ratio_of_sums; mean_of_annual is the alternative and both are always reported. Damodaran's dividends.xls averages ROE over all ten input rows regardless of the selected window, unlike every other statistic in the sheet. Supply extra_years_for_legacy_roe to reproduce that figure when you need to tie out to the sheet, and report the corrected number.

4. The matrix

echo '{"fcfe": 18064.5, "cash_returned": 19869, "roe": 0.0302,
       "cost_of_equity": 0.1122}' | python3 resources/payout.py matrix

Two axes. Can the firm afford its payout, measured by cash returned against FCFE? And does it earn its cost of capital, measured by ROE against the cost of equity?

The insight the matrix protects is that in two quadrants the dividend is not the problem. Recommending a dividend cut in the deficit and poor-projects quadrant, and stopping there, leaves the value destruction untouched. BP cut its dividend by 55% and restructured.

Sequencing matters in the deficit column. With poor projects, fix the investment policy first and then cut. With good projects, reduce the payout so the firm can fund itself rather than issuing equity at a flotation cost.

Optional inputs sharpen the quality axis. Pass roc and cost_of_capital — Damodaran's rule is that ROE above the cost of equity and/or ROC above the cost of capital counts as good projects. Pass jensens_alpha as a cross-check. The engine flags quality_signals_disagree when the signals point different ways.

The trust question is the whole point of the surplus row. A firm with poor returns and a large cash balance should be paying out. Whether it must depends on whether management can be trusted to invest well from here, and that is a judgment about people and governance, not a number. Look at management continuity — Disney's 2003 problem was that the CEO who championed the Capital Cities deal was still in the chair. Ask where the excess returns came from and whether that source persists. Ask whether the board is capable of saying no. Microsoft built a $43bn cash pile by 2002 and shareholders were content, because the project record had earned that trust; Chrysler built an $8bn pile and drew an activist campaign. The arithmetic is identical in both cases. The verdict is not.

State the FCFE variant with the quadrant. The quadrant can change with the variant, and Disney's does.

5. Sustainability

python3 resources/payout.py sustainability --in forecast.json

Assessing the past is half the job. The forward half projects FCFE for five years, projects the dividend the firm will keep paying given stickiness, and takes the difference as repurchase capacity.

ΔWorking Capital_t = WC% × (Revenues_t − Revenues_(t−1))
FCFE_t             = Net Income_t − (Cap Ex_t − Depreciation_t)(1 − DR) − ΔWC_t (1 − DR)
Buyback capacity_t = FCFE_t − Expected Dividends_t

Working capital reinvestment is driven by the revenue increment, not the level. Multiplying the level by the percentage gives the balance and overstates reinvestment by an order of magnitude. The engine does this correctly; check any hand-built model against it.

Growth rates arrive as decimals. The engine rejects anything above 1.0 with an explanation, because entering 5 for 5% is the most common input error here.

Be realistic about the dividend growth rate. Stickiness means the dividend keeps growing at roughly its historical rate unless management commits otherwise. If that produces negative buyback capacity, that is the finding. Do not quietly lower the growth assumption to make the output look better.

Read three fields:

• dividend_sustainable and first_shortfall_year — whether the path holds, and when it

breaks.

• max_sustainable_dividend_growth — the fastest dividend growth this FCFE path covers in

every forecast year. It is the ceiling on what management can promise.

• total_buyback_capacity — the repurchase programme the firm can fund without new

leverage or asset sales.

A payout that cannot be sustained is worse than a lower one. Cuts are read as confessions of failure and are punished accordingly, at roughly −5% to −8% on announcement. Stickiness means firms delay cuts, so an unaffordable dividend can persist for years before a brutal forced cut. Recommending an increase the firm cannot fund in a bad year does real damage. Prefer a buyback for any increase that may not repeat: it returns the same cash with no commitment attached. See resources/methodology.md for the framing evidence on cuts and the constraints — clientele, contractual, regulatory — on executing one.

6. Peers and the market

python3 resources/payout.py peers --in entertainment.json
python3 resources/payout.py market-norms --in disney-2014.json

peers computes dividend yield, dividend payout, cash payout ratio and cash return as a percentage of FCFE for every company, then the group average and median, then the firm's gap and percentile. Averages and medians run over the companies where the metric is defined. A payout ratio on a loss-maker and a cash-return ratio on negative FCFE are both reported as null. Counting them as zero drags both statistics down and is the standard way this table gets misread.